[openib-general] Re: QP error handling

[Sean Hefty]

Roland Dreier wrote:
> mthca currently doesn't handle these 'asynchronous' state transitions
> (ie transition to error).  It continues to think the QP is in the RTS
> state.  Proper handling needs to be implemented.

Ok - thanks for the info.

> However should there be a QP event for a send with invalid L_Key?  I
> would have thought the failed completion entry would be all the
> consumer gets.

I don't think an async event is necessary.  I was working off the failed 
completion entry, but when the modify_qp call failed, I was trying to 
determine if the QP was going into the error state (which would disallow 
the transition) by checking for a callback to the async event handler.

- Sean