Regression: mv88e6xxx packet loss after 4.18's PHYLINK merge

[Samu Nuutamo <samu.nuutamo@vincit.fi>]

Hi,

We have an imx6q-b450v3 board that has one of the switch ports configured as a
fixed link. After upgrading the kernel to version 4.18 the link has experienced
a small amount of packet loss, around 0.15%.

The issue was bisected to a commit: aab9c4067d23 net: dsa: Plug in PHYLINK support

After enabling debug we could see that the new phylink code causes the link to
reset once every second:

[  309.992368] mv88e6085 gpio-0:00 enembc: phylink_mac_config: mode=fixed//100Mbps/Full adv=00000,00000208 pause=10 link=1 an=1
[  309.998451] mv88e6085 gpio-0:00: p5: Force link down
[  309.998869] mv88e6085 gpio-0:00: p5: Speed set to 100 Mbps
[  309.999280] mv88e6085 gpio-0:00: p5: Force full duplex
[  309.999895] mv88e6085 gpio-0:00: p5: Force link up
[  311.032400] mv88e6085 gpio-0:00 enembc: phylink_mac_config: mode=fixed//100Mbps/Full adv=00000,00000208 pause=10 link=1 an=1
[  311.038248] mv88e6085 gpio-0:00: p5: Force link down
[  311.038660] mv88e6085 gpio-0:00: p5: Speed set to 100 Mbps
[  311.039069] mv88e6085 gpio-0:00: p5: Force full duplex
[  311.039678] mv88e6085 gpio-0:00: p5: Force link up
[  312.072328] mv88e6085 gpio-0:00 enembc: phylink_mac_config: mode=fixed//100Mbps/Full adv=00000,00000208 pause=10 link=1 an=1
[  312.077603] mv88e6085 gpio-0:00: p5: Force link down
[  312.078010] mv88e6085 gpio-0:00: p5: Speed set to 100 Mbps
[  312.078417] mv88e6085 gpio-0:00: p5: Force full duplex
[  312.079026] mv88e6085 gpio-0:00: p5: Force link up

It's probably not a coincidence that the link downtime comes out as 0.15%
matching the packet loss percentage. I patched the driver code so that the
reset is skipped if the link state is already set:

--- a/drivers/net/dsa/mv88e6xxx/chip.c
+++ b/drivers/net/dsa/mv88e6xxx/chip.c
@@ -532,6 +532,17 @@ static int mv88e6xxx_port_setup_mac(struct mv88e6xxx_chip *chip, int port,
    if (!chip->info->ops->port_set_link)
        return 0;

+   if (chip->info->ops->port_link_state) {
+       struct phylink_link_state state;
+
+       err = chip->info->ops->port_link_state(chip, port, &state);
+
+       if (!err && link == state.link) {
+           dev_dbg(chip->dev, "link state already set\n");
+           return 0;
+       }
+   }
+
    /* Port's MAC control must not be changed unless the link is down */
    err = chip->info->ops->port_set_link(chip, port, 0);
    if (err)


The patched version did not have any packet loss, but it's probably not the
correct way of fixing the issue.


Best regards,
Samu Nuutamo