[Intel-wired-lan] e1000e Detected Hardware Unit Hang

[Neftin, Sasha <sasha.neftin@intel.com>]

On 3/23/2018 18:09, Marco Berizzi wrote:
> Marco Berizzi wrote:
>>
>> Neftin Sasha wrote:
>>
>>> On 21/03/2018 18:19, Marco Berizzi wrote:
>>>
>>>> Marco Berizzi wrote:
>>>>>
>>>>
>>>>> Hello everyone,
>>>>>
>>>>> I'm getting this error with a Slackware linux 64bit running 4.16-rc6:
>>>>> I have also done a full hardware replacement, but the problem is still
>>>>> present. This box is running squid, named, sshd and strongswan.
>>>>>
>>>>> On february 3, I have replaced the network switch where this NIC is
>>>>> connected to (from cisco 3750 to HPE 5130).
>>>>> As you may see, the problem has popped up from Feb 13 with linux
>>>>> 4.15.3, but yesterday the problem has arisen also with 4.15.2 (I
>>>>> wanted to start a git bisect, but it is pointless).
>>>>
>>>> I think I have found the root cause. Before getting this error the
>>>> interface MTU was 9000bytes, and now it is 1500bytes.
>>>> Good. Please, try also disable TCP segmentation offload: ethtool -K
>>>>
>>>> tso off. Keep us update.
>>
>> Hi Sasha,
>>
>> thanks for the quick response.
>> I have disabled the tso:
>>
>> ethtool -K eth2 tso off
>> Cannot get device udp-fragmentation-offload settings: Operation not supported
>> Cannot get device udp-fragmentation-offload settings: Operation not supported
>>
>> I will keep you updated. Unfortunately I'm encountering this error
>> message not more than once a day. So my response will be a little
>> delayed.
> 
> Hi Sasha,
> 
> after more than 24 hours no error messages has popped up, so this is a
> good workaround.
> 
> Question: is there a way to get the 82579LM working with the tso? Or is this
> an hardware bug?
> 
This is HW bug. Currently we have no workaround in a driver. So, you can 
try work with disable TSO or alternatively play with MTU size. Our more 
new HW should go over this bug.