On Wednesday 19 May 2010 16:35:54 Michael Ellerman wrote:

On Wed, 2010-05-19 at 16:35 +1000, Mark Nelson wrote:

quoted

Hi Michael,

Thanks for looking over these patches!

..

quoted

quoted

Existing code I know, but the error handling in here is a little lax,
what's not going to work if we miss some or all of the interrupts?

That's a good point. For the existing code, if we miss an EPOW event
it just means that the event won't be logged (as that's all we do with
those events at the moment, although there is a comment saying
that it should be fixed to take appropriate action depending upon the
type of power failure); but it's a bigger problem if we miss one of the
RAS errors because then we could miss a fatal event that we should halt
the machine on. And for the upcoming IO events it's even worse as we'd
miss an interrupt from the device...

Yeah that's what I was thinking.

quoted

I would do it in a follow-on patch rather than this one, but what would
be a good course of action if we can't request the interrupt?

Yes a follow on patch is the way to do it.

There shouldn't be that many reasons the request fails, other than
ENOMEM, or broken device tree perhaps. It's definitely worth a
WARN_ON(), people notice those at least.

That sounds good. I'll do a simple follow-on patch that adds the
WARN_ON().

Thanks!
Mark