Can Intel 82574L (em) turn off flow control?

[Lev Serebryakov]

 I'm seeing very strange situation when try to send a lot of traffic
(with netmap pkt-gen) from 82574L to 82574L NICs.

 I'm using 13-CURRENT, with fc turned off (dev.em.X.fc=0). But still,
`pkt-gen` can not send more packets per second than receiving side could
process!

  When 82574L NIC sends to i210 NIC it could send much more than
receiving side could process, and there are expected number of lost packets.

 But when receiving side is 82574L too, `pkt-gen -f tx` shows statisitcs
like this:

# pkt-gen -f tx -N -4 -i em1 -l 60 -R 744000 <skipped>
...
531.511259 main_thread [2464] 217104 pps (218624 pkts 104939520 bps in
1007000 usec) 429.52 avg_batch 99999 min_space
532.547259 main_thread [2464] 216958 pps (224768 pkts 107888640 bps in
1035999 usec) 426.50 avg_batch 99999 min_space
533.548256 main_thread [2464] 216872 pps (217088 pkts 104202240 bps in
1000998 usec) 426.50 avg_batch 99999 min_space
...

 217104 pps instead of requested 744000! Generator could generate much
more if I have, say, switch on other side.

 Is it limitation of 82574L or bug in driver?

-- 
// Lev Serebryakov

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 963 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/freebsd-net/attachments/20190226/ed6f7b0d/attachment.sig>