On Mon, 2008-12-08 at 09:58 -0800, Sean Hefty wrote: > >Since the IBTA 1.2.1 spec. clarified that the RESET to ERROR QP state > >transition is not valid but earlier the openfabrics code supported it, > >the code in cma_modify_qp_err() will now return an error if the QP > >is in the RESET state. This can cause RDS to go into a loop trying to > >call rdma_disconnect() continuously. > > Can you explain when the QP is in the RESET state when cma_modify_qp_err() is > called? > > - Sean
On closer inspection, I think this patch isn't needed. It looks like ib_modify_qp() to INIT is always called right after creating the QP and if there is an error, the QP is destroyed so it shouldn't be possible for the QP to be left in the RESET state. I was debugging an RDS problem and seeing rdma_disconnect() returning errors but now I'm not sure why I thought it was due to the ib_modify_qp() to ERR that was failing. _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
