Our CI testing has been failing for NFS RDMA since 5.14-rc1.
Based on kprobes, the NFS RDMA client creates its QP using rdma_create_qp(), and does post receives right away.
This patch below looks like it deleted the transition from RESET to INIT, breaking the client side NFS RDMA since post receives are not valid in RESET.
I suspect the patch needs to be reverted or NFS RDMA needs to handle the transition to INIT?
commit dc70f7c3ed34b081c02a611591c5079c53b771b8
Author: H�kon Bugge <haakon.bugge@xxxxxxxxxx>
Date: Tue Jun 22 15:39:56 2021 +0200
RDMA/cma: Remove unnecessary INIT->INIT transition
In rdma_create_qp(), a connected QP will be transitioned to the INIT
state.
Afterwards, the QP will be transitioned to the RTR state by the
cma_modify_qp_rtr() function. But this function starts by performing an
ib_modify_qp() to the INIT state again, before another ib_modify_qp() is
performed to transition the QP to the RTR state.
Hence, there is no need to transition the QP to the INIT state in
rdma_create_qp().
Link: https://lore.kernel.org/r/1624369197-24578-2-git-send-email-haakon.bugge@xxxxxxxxxx
Signed-off-by: H�kon Bugge <haakon.bugge@xxxxxxxxxx>
Reviewed-by: Leon Romanovsky <leonro@xxxxxxxxxx>
Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
