Re: [dpdk-dev] [PATCH] net/bnxt: fix regression in non-vector mode Rx data path on Thor

[Ajit Khaparde <ajit.khaparde@broadcom.com>]

On Wed, Nov 3, 2021 at 9:10 AM Somnath Kotur <somnath.kotur@broadcom.com> wrote:
>
> The patch introduced by
> commit 657c2a7f1dd4 ("net/bnxt: create aggregation rings when needed")
> ended up shortening the return code path from the function thereby
> resulting in not executing the line of code at the end of the function
> that was resetting the next consumer index to 0.
> This would result in an application crash when error recovery or other
> port stop/start scenarios were invoked on Thor which is what this
> commit 61cd4384fabf ("net/bnxt: fix crash after port stop/start")
> was addressing.
> Fix it by moving the resetting line of code before the return path
> in the case when aggregration rings are not used (default case).
>
> Fixes: 657c2a7f1dd4 ("net/bnxt: create aggregation rings when needed")
>
> Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Fixed typo in commit log.
Patch applied to dpdk-next-net-brcm.

> ---
>  drivers/net/bnxt/bnxt_rxr.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/net/bnxt/bnxt_rxr.c b/drivers/net/bnxt/bnxt_rxr.c
> index 992f97388c..8bc8ddc353 100644
> --- a/drivers/net/bnxt/bnxt_rxr.c
> +++ b/drivers/net/bnxt/bnxt_rxr.c
> @@ -1369,6 +1369,9 @@ int bnxt_init_one_rx_ring(struct bnxt_rx_queue *rxq)
>                 rxr->rx_buf_ring[i] = &rxq->fake_mbuf;
>         }
>
> +       /* Explicitly reset this driver internal tracker on a ring init */
> +       rxr->rx_next_cons = 0;
> +
>         if (!bnxt_need_agg_ring(rxq->bp->eth_dev))
>                 return 0;
>
> @@ -1407,9 +1410,6 @@ int bnxt_init_one_rx_ring(struct bnxt_rx_queue *rxq)
>         }
>         PMD_DRV_LOG(DEBUG, "TPA alloc Done!\n");
>
> -       /* Explicitly reset this driver internal tracker on a ring init */
> -       rxr->rx_next_cons = 0;
> -
>         return 0;
>  }
>
> --
> 2.28.0.497.g54e85e7
>