. > > I suspect the patch needs to be reverted or NFS RDMA needs to handle the > transition to INIT? > Reverting the patch below works. > 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 A quick audit of .post_recv calls: 1 main.c <global> 698 .post_recv = bnxt_re_post_recv, <-- allows 2 provider.c <global> 489 .post_recv = c4iw_post_receive, <-- allows 3 hns_roce_hw_v1.c <global> 4411 .post_recv = hns_roce_v1_post_recv, <-- allows 4 hns_roce_hw_v2.c <global> 6190 .post_recv = hns_roce_v2_post_recv, <-- -EINVAL 5 verbs.c <global> 4396 .post_recv = irdma_post_recv, <-- allows 6 main.c <global> 2561 .post_recv = mlx4_ib_post_recv, <-- allows 7 main.c <global> 3767 .post_recv = mlx5_ib_post_recv_nodrain, <-- allows 8 mthca_provider.c <global> 1148 .post_recv = mthca_arbel_post_receive, <-- allows 9 mthca_provider.c <global> 1154 .post_recv = mthca_tavor_post_receive, <-- allows a ocrdma_main.c <global> 173 .post_recv = ocrdma_post_recv, <-- -EINVAL b main.c <global> 221 .post_recv = qedr_post_recv, <- -EINVAL c pvrdma_main.c <global> 175 .post_recv = pvrdma_post_recv, <- -EINVAL d vt.c <global> 392 .post_recv = rvt_post_recv,<- -EINVAL e rxe_verbs.c <global> 1132 .post_recv = rxe_post_recv, <- -EINVAL f siw_main.c <global> 287 .post_recv = siw_post_receive, <-- allows? Looks like it is a non-portable assumption that an out-of-state post recv works. I guess it is possible that the post happens and gets flushed back? >From the IBTA spec: Reset to Init - Enable posting to the Receive Queue ... post receive notes "Invalid QP state" as a possible immediate failure, but nothing more binding than that. Mike
