mtk_eth_soc MT7988: "warm reset failed" → page_pool stuck → permanent TX wedge

[danielschonfeld]

6.10-main · BPI-R4 4GB · GPON SFP on eth2 (inband/1000base-x, 1G)

A link event triggers a warm reset that fails; page_pool id 6 then can't release and TX wedges on both eth0 and eth2 until a hard power cycle. Recurs every few weeks.

mtk_soc_eth 15100000.ethernet: warm reset failed
page_pool_release_retry() stalled pool shutdown: id 6, 7498 inflight 60 sec
mtk_soc_eth 15100000.ethernet eth0: NETDEV WATCHDOG: transmit queue 4 timed out   (repeats ~5s, forever)

Box stays up (CPU/userspace fine); only 15100000.ethernet is dead — ip link/reboot don't help, power cycle only. Not the lynxi link_up reconfig (absent in 6.10). Not ISP-side.

Log excerpt (Aug 30)

Aug 30 13:32:54 mtk_soc_eth 15100000.ethernet: warm reset failed
Aug 30 13:32:54 eth0: configuring for fixed/internal link mode
Aug 30 13:32:54 eth2: configuring for inband/1000base-x link mode
Aug 30 13:32:54 eth0: Link is Up - 10Gbps/Full
Aug 30 13:32:54 eth2: Link is Up - 1Gbps/Full
Aug 30 13:32:59 eth0: NETDEV WATCHDOG: transmit queue 4 timed out
Aug 30 13:33:24 eth2: NETDEV WATCHDOG: transmit queue 2 timed out
Aug 30 13:33:54 page_pool_release_retry() stalled pool shutdown: id 6, 7498 inflight 60 sec
Aug 30 13:34:54 page_pool_release_retry() stalled pool shutdown: id 6, 7293 inflight 120 sec
(eth2 flaps Up/Down throughout; watchdog repeats ~5s until power cycle)

[frank-w]

could you check if newer kernel version works better? but the transmit timeouts are hard to trace. i'm afraid i cannot help here much

[danielschonfeld]

the newer kernels suffer from issue #185 for me. is there any progress on that front?