[dpdk-stable] [PATCH v2 2/5] app/testpmd: enable burst stats for flowgen mode rx path

[Honnappa Nagarahalli]

<snip>

> >
> >> Subject: Re: [PATCH v2 2/5] app/testpmd: enable burst stats for
> >> flowgen mode rx path
> >>
> >> On 6/26/2020 11:09 PM, Honnappa Nagarahalli wrote:
> >>> Enable missing burst stats for rx path for flowgen mode.
> >>>
> >>> Fixes: e9e23a617eb8 ("app/testpmd: add flowgen forwarding engine")
> >>> Cc: stable at dpdk.org
> >>>
> >>> Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli at arm.com>
> >>> Reviewed-by: Phil Yang <phil.yang at arm.com>
> >>> Reviewed-by: Ruifeng Wang <ruifeng.wang at arm.com>
> >>> Tested-by: Phil Yang <phil.yang at arm.com>
> >>> ---
> >>>  app/test-pmd/flowgen.c | 4 ++++
> >>>  1 file changed, 4 insertions(+)
> >>>
> >>> diff --git a/app/test-pmd/flowgen.c b/app/test-pmd/flowgen.c index
> >>> 4bd351e67..011887c61 100644
> >>> --- a/app/test-pmd/flowgen.c
> >>> +++ b/app/test-pmd/flowgen.c
> >>> @@ -111,6 +111,10 @@ pkt_burst_flow_gen(struct fwd_stream *fs)
> >>>  	/* Receive a burst of packets and discard them. */
> >>>  	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
> >>>  				 nb_pkt_per_burst);
> >>> +#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
> >>> +	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
> >>> +#endif
> >>> +
> >>
> >> Flow generator is for Tx, what is the point of getting stats for the Rx?
> > Is there any reason for this mode to receive the packets?
> > The reason I ask is, the core cycles include the RX call as well. So, we have
> cycles/packet that includes the RX call and burst stats that does not show the
> RX burst stats.
> > Do we need to remove the call to rte_eth_rx_burst API?
> >
> 
> The implementation receives packets and frees them immediately.
> I assume the Rx is implemented in case port receives packets, like if other end
> forwards the generated packets back to same port.
I tried to find more info, there is not a lot. The original commit message says 'it is similar to 'txonly' engine, but will generate multiple L4 flows'.
Your assumption seems reasonable to me. In that case, I think it makes sense to include 'core cycles' only for TX. Otherwise, there is not enough information to debug any issues.

> Yes it consumes cycles for Rx but received packets are ignored, I am not sure if
> received packets have any meaning in flowgen to get additional burst stats.
Please see above, I think we need to adjust how the 'core cycles' are counted in this case.