On Mon, Aug 2, 2021 at 12:59 PM Leon Romanovsky <leon@xxxxxxxxxx> wrote: > > On Mon, Aug 02, 2021 at 11:37:06AM -0400, Olga Kornievskaia wrote: > > Hi folks, > > > > We are encountering an error condition (while doing NFSoRDMA) but the > > problem seems to be in the RDMA core itself. The problem is that the > > client at some point is ending in an RDMA NAK with "PNS Sequence > > error" but the network trace shows all the PSNs are accounted for > > (snippet at the bottom). It's as if the client lost its knowledge of > > the current PSN. > > > > Questions: > > 1. Is PSN handling done by the hardware card itself (in firmware) and > > not in the kernel (making this a card/firmware specific problem)? I > > was trying to look thru the rdma core/mlx5 driver code to see what > > would generate a NAK with such error but wasn't able to find one. Only > > found counters for nak_seq_error which made me think this is a > > firmware problem. > > The decisions what is valid or not are done in the FW, kernel doesn't > check anything. Although, the kernel sets/gets next_send_psn/next_recv_psn. Thank you for the confirmation. > > Thanks
