From: Kishore Padmanabha <[email protected]> The packets are getting dropped if some of the queues are configured to stop and if the bnxt driver is reset due to port stop all and port start all then when the bnxt driver comes up, the driver is overwriting the queue configuration to operationally started instead of stopping the queues that was previously stopped. The stopped queues were getting added to the rss context resulting in the packet drops.
Added a fix to address this issue, the rxq structure is not cleared during the stop and start process and retains the context, that flag is used to set the queue state. Thus stopped queues shall not be added to the rss context on recovery from a stop and start. The queues that were configured to be stopped have to explicitly started to see traffic on those queues. Signed-off-by: Kishore Padmanabha <[email protected]> Reviewed-by: Somnath Kotur <[email protected]> Reviewed-by: Peter Spreadborough <[email protected]> Reviewed-by: Kalesh AP <[email protected]> Reviewed-by: Ajit Khaparde <[email protected]> --- drivers/net/bnxt/bnxt_ethdev.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c index efecbba2c3..c92e335a51 100644 --- a/drivers/net/bnxt/bnxt_ethdev.c +++ b/drivers/net/bnxt/bnxt_ethdev.c @@ -925,11 +925,13 @@ static int bnxt_start_nic(struct bnxt *bp) for (j = 0; j < bp->rx_nr_rings; j++) { struct bnxt_rx_queue *rxq = bp->rx_queues[j]; + __rte_assume(j < RTE_MAX_QUEUES_PER_PORT); + /* If not deferred start then change only the state of the */ + /* queue based on the queue rx_started flag */ if (!rxq->rx_deferred_start) { - __rte_assume(j < RTE_MAX_QUEUES_PER_PORT); - bp->eth_dev->data->rx_queue_state[j] = + if (rxq->rx_started) + bp->eth_dev->data->rx_queue_state[j] = RTE_ETH_QUEUE_STATE_STARTED; - rxq->rx_started = true; } } -- 2.39.5 (Apple Git-154)
