Account for rx_dropped packets in linux ifdropped

[gtedesco-r7]

Many Linux drivers are not reporting anything into rx_missed_errors or rx_fifo_errors. They, instead, report dropped packets into rx_dropped.

Without including rx_dropped into ps_ifdrop, we can get a wildly optimistic view of performance.

This particularly affects vmxnet3, enic, ena, cxgb, and cxgb4 drivers.

Issue described in: #1328

[infrastation]

Thank you for looking into this problem. In the drivers you mention only Chelsio is a hardware. Do you know if the same logic stands for common 1GbE interfaces found in desktops and servers? The drivers in kernel/drivers/net/ethernet/(broadcom|intel|realtek) would be good examples.

[gtedesco-r7]

Yeah, so intel ice for example uses rx_dropped and not rx_missed. i40e uses both rx_missed and rx_dropped. Ah but those are 10/40gbe.

[gtedesco-r7]

tg3 driver uses rx_dropped for aggregated per-rxq drop stats, and rx_missed_errors only includes a global counter, for example

[gtedesco-r7]

All the really old, classic, drivers like realtek 81392 and tulip use rx_missed_errors only, so maybe that was the only stat that existed back in the days of "pump up the jam" and fresh prince of bel air :)