On 9/19/2018 8:55 PM, Dan Gora wrote:
> The worker threads incrementing the rx/tx_packets race with the signal
> handler from the main thread zeroing the entire statistics structure.
> This can cause the statistics to fail to be zeroed, even when there
> is no traffic on those interfaces.
>
> Improve zeroing the statistics by only incrementing rx/tx_packets
> in worker threads by a non-zero amount. This limits the race to the
> periods in which traffic is actually being received or transmitted.
Not sure about introducing an extra check to datapath for possible error on
stats zero. I am for dropping this patch, what do you think?
>
> Signed-off-by: Dan Gora <d...@adax.com>
> ---
> examples/kni/main.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/examples/kni/main.c b/examples/kni/main.c
> index ca45347d8..d00569740 100644
> --- a/examples/kni/main.c
> +++ b/examples/kni/main.c
> @@ -223,7 +223,8 @@ kni_ingress(struct kni_port_params *p)
> }
> /* Burst tx to kni */
> num = rte_kni_tx_burst(p->kni[i], pkts_burst, nb_rx);
> - kni_stats[port_id].rx_packets += num;
> + if (num)
> + kni_stats[port_id].rx_packets += num;
>
> rte_kni_handle_request(p->kni[i]);
> if (unlikely(num < nb_rx)) {
> @@ -260,7 +261,8 @@ kni_egress(struct kni_port_params *p)
> }
> /* Burst tx to eth */
> nb_tx = rte_eth_tx_burst(port_id, 0, pkts_burst, (uint16_t)num);
> - kni_stats[port_id].tx_packets += nb_tx;
> + if (nb_tx)
> + kni_stats[port_id].tx_packets += nb_tx;
> if (unlikely(nb_tx < num)) {
> /* Free mbufs not tx to NIC */
> kni_burst_free_mbufs(&pkts_burst[nb_tx], num - nb_tx);
>
