[openib-general] [PATCH 3/5 v3] 2.6.20 rdma/cma: allow early transition to RTS to handle lost CM messages

[Michael S. Tsirkin]

> Michael S. Tsirkin wrote:
> >> To handle the case
> >> where the connection messages are lost, a new API is added that users
> >> may invoke to force a connection into the established state.
> > 
> > Just to clarify this point - what connecton messages can be lost?
> > E.g. if the passive side does not get an RTU for a while, it will
> > retry the REP, won't it?  Diagram 12.9.6 seems to indicate so:
> > from REP Sent we should go to RTU timeout, Send REP and back to REP Sent.
> > Is this implemented?
> 
> It handles the case where the first RX crosses the RTU which can happen 
> when the RTU is lost but also without it being lost.
> 
> Indeed the passive side would resend the REP when a timeout expires but 
> the patch allows the app to force the connection establishment **now** 
> (ie have the CMA move the RC QP to RTS) and not go into queuing of RX-es 
> etc till the RTU is lost, it also handles the case where all the RTUs 
> are lost.

I think we all already agreed we need the rdma_established call,
for reasons that you outline. So I am not arguing at all - I was just
checking that REP re-sends are implemented.

So, a slightly more exact description for the patch would be
"to handle the case where a data packet bypasses an RTU".
Is that right?

-- 
MST