ramips: gsw_mt7621: disable PORT 5 MAC RX/TX flow control by default
authorPetr Štetiar <ynezz@true.cz>
Tue, 11 Feb 2020 10:17:41 +0000 (11:17 +0100)
committerFelix Fietkau <nbd@nbd.name>
Thu, 20 Feb 2020 14:08:06 +0000 (15:08 +0100)
commitc8f8e59816eca49d776562d2d302bf990a87faf0
tree0cb15a8f86e08ddf9eed49f6719b4425072b27f2
parent3e11ddaf2ede4f105bc9ac91229623526371a7a2
ramips: gsw_mt7621: disable PORT 5 MAC RX/TX flow control by default

Looking at the current upstream driver implementation, it seems like the
TX/RX flow control is enabled only if the flow control pause option is
resolved from the device/link partner advertisements (or otherwise set).

On the other hand, our current in-tree driver force enables TX/RX
flow control by default, thus possibly leading to TX timeouts if the
other end sends pause frames (which are not properly handled?):

 WARNING: CPU: 3 PID: 0 at net/sched/sch_generic.c:320 dev_watchdog+0x1ac/0x324
 NETDEV WATCHDOG: eth0 (mtk_soc_eth): transmit queue 0 timed out

Disabling the flow control on PORT 5 MAC seems to fix this issues as the
pause frames are then filtered out. While at it, I'm removing the if
condition completely as suggested, since this code is run only on mt7621
SoC, so there is no need to check for the silicon revisions.

Ref: https://lists.openwrt.org/pipermail/openwrt-devel/2017-November/009882.html
Ref: https://forum.openwrt.org/t/mtk-soc-eth-watchdog-timeout-after-r11573/50000/12
Suggested-by: Felix Fietkau <nbd@nbd.name>
Reported-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: Petr Štetiar <ynezz@true.cz>
target/linux/ramips/files-4.14/drivers/net/ethernet/mediatek/gsw_mt7621.c