[Intel-wired-lan] igb Detected Tx Unit Hang after upgrade to 4.18-rc6 [was Re: igb Detected Tx Unit Hang after upgrade to 4.17]

[Marco Berizzi]

> Il 1 settembre 2018 alle 4.01 Alexander Duyck <alexander.duyck at gmail.com> ha scritto:

> However with that being said it does occur to me that you mentioned
> using VirtualBox. I dug back through the dmesg output and found the
> line:
> [   20.944982] VBoxNetFlt: attached to 'eth0' / 00:19:99:ab:0b:38
> 
> I think there may be an issue with the filter stripping out packets at
> the end of a queue and as a result we are receiving packets with
> xmit_more set and not updating tail. At least that is my understanding
> of things for now as I am not that familiar with the VBoxNetFlt
> driver.
> 
> There are two ways to verify if this is the case. The first and more
> correct solution might be to look at updating
> vboxNetFltLinuxStartXmitFilter so that it will clear the xmit_more bit
> before handing it off to the lower device. If you don't have access to
> that an alternative would be to just modify the igb driver you have so
> that you replace "skb->xmit_more" with "true" and just force the tail
> to always be written out. It will cause a small performance hit but it
> will work around the issue in the driver that overwrote our network
> device ops.
> 
> Hope that helps.
> 
> - Alex

Thanks a lot Alexander for debugging this problem.
I'm confident the virtualbox people will take care
of this problem.

Thanks again for your prompt response.