Paul Mackerras paulus at
Wed Mar 24 13:23:47 EST 2004

Greg KH writes:

> On Tue, Mar 23, 2004 at 11:32:07AM +1100, Paul Mackerras wrote:
> > I'm wondering what you mean by "before anything touches the card"
> > though.  The driver will be in the middle of doing things with the
> > card when the EEH event happens, so it will be touching the card until
> > we get the notification to it that the event has happened.
> I mean, "before the power is yanked away from the card", which is what
> you all are doing today (disable the power, and as that happens the
> hotplug event happens.)

Well, power doesn't get yanked away from the card.  The card is cut
off from the processor by the PCI-PCI bridge, but that has already
happened -- the hardware did that some small number of nanoseconds
after it detected an error.  We have no control over that.

The card remains powered on, just inaccessible.  We can reset the slot
and then reconnect it, using firmware (RTAS) calls.  Whether or not we
do that is, or how many times we do that before giving up, is a policy
we can decide.  But the initial disconnection isn't policy -- or at
least isn't policy that we can decide.  By the time we know that
anything has gone wrong, the disconnection has already happened.


** Sent via the linuxppc64-dev mail list. See

More information about the Linuxppc64-dev mailing list