Merge branch 'mediatek-various-performance-improvements'
authorDavid S. Miller <davem@davemloft.net>
Tue, 20 Jun 2017 17:40:36 +0000 (13:40 -0400)
committerDavid S. Miller <davem@davemloft.net>
Tue, 20 Jun 2017 17:40:36 +0000 (13:40 -0400)
commitc036cab4ee7ba946a29ce0bdce5ad758347a11d2
tree8be18b2ff9e19d03a322acf2dc73e467fc73ac6a
parentcf67bebecb4b5564586a4674646382ca380ec205
parenta2d5e7b4102deac784373464a8fd9f3eaa53afc0
Merge branch 'mediatek-various-performance-improvements'

John Crispin says:

====================
net-next: mediatek: various performance improvements

During development we mainly ran testing using iperf doing 1500 byte
tcp frames. It was pointed out recently, that the driver does not perform
very well when using 512 byte udp frames. The biggest problem was that
RPS was not working as no rx queue was being set. fixing this more than
doubled the throughput. Additionally the IRQ mask register is now locked
independently for RX and TX. RX IRQ aggregation is also added. With all
these patches applied we can almost triple the throughput.

While at it we also add PHY status change reporting for GMACs connecting
directly to a PHY.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>