Bug#932996: linux-image-4.19.0-5-amd64: e1000e driver crashes under high TX load; `ethtool -K eno1 tso off` resolves issue.

[John <john@narbondel.org>]

Hi,

I've turned TSO back on to test, and I'm currently on 5.10.0-7-amd64 ($ uname -a -> Linux myrkul 5.10.0-7-amd64 #1 SMP Debian 5.10.40-1 (2021-05-28) x86_64 GNU/Linux).

I could not readily reproduce it using a netcat tunnel to push higher load from VM to host, or from VM to external networks, so it looks likely that something in the last two years cleaned this up. Feel free to close, and I can reopen with more details if it comes back.

Thanks!

- John

On Sun, Jun 13, 2021 at 12:11 PM Salvatore Bonaccorso <carnil@debian.org> wrote:

Control: tags -1 + moreinfo

On Thu, Jul 25, 2019 at 09:58:33AM -0700, John wrote:
> Package: src:linux
> Version: 4.19.37-6
> Severity: normal
>
> Dear Maintainer,
>
> Observed flaky network interface under high TX load with guest OS
> bridging across eno1 interface. Issue only manifested under load, and
> exhibited flapping in dmesg, e.g.:
> e1000e 0000:00:1f.6 eno1: Reset adapter unexpectedly
> br0: port 1(eno1) entered disabled state).
>
> Issue appears to be resolved by disabling TSO with:
> `ethtool -K eno1 tso off`
>
> Googling around suggests this may be a common problem with this driver
> -- possibly should default to tso off? See possibly related links:
> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1766377
> https://serverfault.com/questions/616485/e1000e-reset-adapter-unexpectedly-detected-hardware-unit-hang

Is this issue still triggerable for you with a recent kernel from
either unstable, buster-backports or even mainline without needing to
set tso off?

Regards,
Salvatore