Public bug reported:
Users are reporting that the OOB network occasionally becomes
unreachable on BlueField 4. Investigation has shown that when this issue
happens TX completion interrupts from the LAN7430 NIC are no longer
received, effectively causing a TX hang. The kernel's TX queue fills and
TX counters stop incrementing. The only way to recover is by manually
taking the interface down and then bringing it back up.
A root cause has not been identified yet, but we verified that in this
TX hang case the kernel's netdev watchdog will run if supported and can
be used to automatically recover.
This patch series adds the following support to the LAN743x kernel
driver:
1. Implement the ndo_tx_timeout handler to effectively bring the interface down
and up again to recover after a TX hang is detected by the kernel.
2. Add BQL support to decrease TX watchdog trigger time. Now instead of waiting
for the entire TX queue to be full before triggering the watchdog the kernel
can trigger based on the difference of bytes in the TX queue and bytes that
have been transmitted.
3. Stability and error checking improvements related to the close/open flows.
** Affects: linux-nvidia-bos (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2170884
Title:
lan743x: automatically recover from TX hang
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia-bos/+bug/2170884/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
