dpaa2-eth: Don't use netif_receive_skb_list for TCP frames
authorIoana Radulescu <ruxandra.radulescu@nxp.com>
Mon, 22 Jul 2019 15:52:39 +0000 (18:52 +0300)
committerDong Aisheng <aisheng.dong@nxp.com>
Mon, 14 Dec 2020 03:36:29 +0000 (11:36 +0800)
commitb431cb87be02dc69d1c3028564fe34869a53d038
tree6e29609faeea75d2604301f19ecf29b15ded1330
parent03d359ce7c7aa8b3f56e8fd614597b6c5b659cff
dpaa2-eth: Don't use netif_receive_skb_list for TCP frames

Using Rx skb bulking for all frames may negatively impact the
performance in some TCP termination scenarios, as it effectively
bypasses GRO.

Look at the hardware parse results of each ingress frame to see
if a TCP header is present or not; for TCP frames fall back to
the old implementation.

Signed-off-by: Ioana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h