[ofa-general] [PATCH v2] RDMA/cxgb3: Handle EEH events for active connections.
Steve Wise
swise at opengridcomputing.com
Fri Mar 6 15:46:39 PST 2009
Roland Dreier wrote:
> > - post QP_FATAL event on all active QPs when cxgb3 notifies
> > iw_cxgb3 of a fatal error.
>
> This actually doesn't match what IB adapters do: if the whole adapter is
> being reset, mlx4 and mthca just report DEVICE_FATAL for the whole
> device. I think the QP_FATAL event should be when something gets
> screwed up so some work requests are lost (ie no completions will ever
> come) but the device as a whole is OK.
>
> As an example, the Open MPI code seems to handle DEVICE_FATAL but not do
> the same thing for QP_FATAL etc.
>
> - R.
>
My last version of this patch posted DEVICE_FATAL and you said it wasn't
right.
I'm confused.
More information about the general
mailing list