SFP ports stop working with Linux 6.14 in Arch Linux

Ten64 Forum
1

When upgrading my box today to the latest kernel, I found that I lost all connectivity on the SFP ports. When I rolled back to kernel version 6.13, they worked again.

On the bad kernel version, I do see these log lines:

Apr 10 19:20:05 Heimdall kernel: fsl_dpaa2_eth dpni.1 (unnamed net_device) (uninitialized): autoneg setting not compatible with PCS
Apr 10 19:20:05 Heimdall kernel: sfp dpmac2-sfp: Unable to ascertain link mode
Apr 10 19:20:05 Heimdall kernel: fsl_dpaa2_eth dpni.1 eth8: selection of interface failed, advertisement 00,00000000,00000000,00000000

When I have time, I’ll likely do a bisect to locate the exact commit if I can.

2

This appears to be due to a feature introduced in the 6.14 cycle, which deals with in-band negotiation (something generally used with <=1Gbps copper ports and SFPs)

https://patchwork.kernel.org/project/netdevbpf/cover/Z08kCwxdkU4n2V6x@shell.armlinux.org.uk/

And specifically this patch to the PCS driver used on this hardware (lynx):
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v6.15-rc1&id=6561f0e547be221f411fda5eddfcc5bd8bb058a5

I have just tested reverting the above commit on the Linus tree and the SFPs work again.

I’ll do some more checking next week, before formally reporting it as a regression.

2 Likes
3

Russell King has posted this patch to try:

diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
index 1bdd5d8bb5b0..2147e2d3003a 100644
--- a/drivers/net/phy/phylink.c
+++ b/drivers/net/phy/phylink.c
@@ -3624,6 +3624,15 @@ static int phylink_sfp_config_optical(struct phylink *pl)
 	phylink_dbg(pl, "optical SFP: chosen %s interface\n",
 		    phy_modes(interface));
 
+	/* GBASE-R interfaces with the exception of KR do not have autoneg at
+	 * the PCS. As the PCS is media facing, disable the Autoneg bit in the
+	 * advertisement.
+	 */
+	if (interface == PHY_INTERFACE_MODE_5GBASER ||
+	    interface == PHY_INTERFACE_MODE_10GBASER ||
+	    interface == PHY_INTERFACE_MODE_25GBASER)
+		__clear_bit(ETHTOOL_LINK_MODE_Autoneg_BIT, config.advertising);
+
 	if (!phylink_validate_pcs_inband_autoneg(pl, interface,
 						 config.advertising)) {
 		phylink_err(pl, "autoneg setting not compatible with PCS");

The SFPs are now working again on my system. Can you check if the above works for you as well?

4

Can confirm that that patch on top of commit bc3372351d0c8b2726b7d4229b878342e3e6b0e8 fixes the SFP ports.

1 Like