Symptom
Handing the tempo-master role back to a deck sometimes drops the Ethernet
link for a couple of seconds:
[bridge] device-number claim complete for device 7
[eth] link down
[bridge] keepalive → device=7 FAILED
[bridge] stop (silence timeout)
[eth] link up
[bridge] waiting for downbeat (got beat 4/4)
[bridge] start (downbeat)
It is a genuine ETHERNET_EVENT_DISCONNECTED from the W5500 driver, not a
software artifact, and there is no reboot (uptime keeps climbing across it).
What the captures show
Across two ~10-minute serial captures:
| Transition |
Link drop |
| Release-side claim (idle device 7) |
3 of 4 |
| Take-master claim (device 4) |
0 of 4 |
| Any other time |
never |
Strongly correlated with the release-side claim, but not deterministic.
Impact
Mitigated, not fixed. The clock used to send MIDI Stop and restart on the
next downbeat; it now rides out a known-down link for link_down_grace_ms
(8 s) and free-runs on the latched tempo, so the drop is no longer audible.
The link still drops.
Leading suspicion
Power. The claim burst is 12 broadcasts in ~600 ms, and a TX current spike
on a marginal supply could reset the PHY — which would explain why it is
probabilistic. The unit was running off a laptop USB port during these
captures, and this build has a history of power sensitivity.
Next steps
- Reproduce on a proper wall supply vs. laptop USB.
- Scope the 3V3 rail across the claim burst.
- If it is not power: instrument the W5500 driver (raise its log level in the
diag build) to see what the driver believes happened.
Symptom
Handing the tempo-master role back to a deck sometimes drops the Ethernet
link for a couple of seconds:
It is a genuine
ETHERNET_EVENT_DISCONNECTEDfrom the W5500 driver, not asoftware artifact, and there is no reboot (uptime keeps climbing across it).
What the captures show
Across two ~10-minute serial captures:
Strongly correlated with the release-side claim, but not deterministic.
Impact
Mitigated, not fixed. The clock used to send MIDI Stop and restart on the
next downbeat; it now rides out a known-down link for
link_down_grace_ms(8 s) and free-runs on the latched tempo, so the drop is no longer audible.
The link still drops.
Leading suspicion
Power. The claim burst is 12 broadcasts in ~600 ms, and a TX current spike
on a marginal supply could reset the PHY — which would explain why it is
probabilistic. The unit was running off a laptop USB port during these
captures, and this build has a history of power sensitivity.
Next steps
diagbuild) to see what the driver believes happened.