On 2026-09-24, Matt Vollrath wrote:
> I didn't try your patch, but there's a four line fix. Where the watchdog
> resets just the PHY when the MAC exits DMoff state, set FLAG_RESTART_NOW
> instead of resetting the PHY, then goto link_up. This works for me at
> both 1500 and 9000 MTU.
Thanks Matt,
I've tested this approach on two I219-LM boxes, each attached to a different
switch:
one a TP-Link TL-SG108PE V3, and the other a Cisco Catalyst 9200.
On the Cisco with jumbo frames, I'm seeing, via a diagnostic message, that
the fix appears to fire and work. I have also noticed that I no longer see
it on every boot, but this may be correct as I've not seen traffic stalls.
With 1500 mtu, I'm seeing it engage fairly regularly with the TP-Link but
not the Cisco. I'm guessing it is related to how the Cisco negotiates. In both
cases
traffic doesn't seem to stall.
I think you are correct about flow control being a red herring since
both switches advertise no pause (Flow Control: None) before and after
recovery.
I'd like to drop my patch in favour of your fix.
How should I proceed, should I propose a patch with suggested-by you, or with
your authorship and
tested by me?
For completeness, here is the diff I tested (I added an e_info so I could see
it fire on the
test rig; happy to drop that line - it isn't needed for the fix):
--- a/drivers/net/ethernet/intel/e1000e/netdev.c
+++ b/drivers/net/ethernet/intel/e1000e/netdev.c
@@ -5254,9 +5254,16 @@
usleep_range(10000, 20000);
pcim_state = er32(STATUS);
- /* Checking if MAC exited DMoff state */
- if (!(pcim_state &
E1000_STATUS_PCIM_STATE))
-
e1000_phy_hw_reset(&adapter->hw);
+ /* Checking if MAC exited DMoff state.
A bare
+ * PHY reset here does not re-run the
jumbo
+ * workaround
(e1000_lv_jumbo_workaround_ich8lan),
+ * so request a full reset instead,
which does.
+ */
+ if (!(pcim_state &
E1000_STATUS_PCIM_STATE)) {
+ e_info("full reset on DMoff
exit to restore jumbo-frame configuration\n");
+ adapter->flags |=
FLAG_RESTART_NOW;
+ goto link_up;
+ }
}
}
