Messages in this thread | | | From | Holger Hoffstätte <> | Subject | Re: 1e918876 breaks r8169 (linux-3.18+) | Date | Sun, 8 Feb 2015 12:18:33 +0000 (UTC) |
| |
On Fri, 06 Feb 2015 15:04:50 +0100, Tomas Szepe wrote:
> Unfortunately, I have to take this back. I made the conclusion too early. > The problem appears with this patch applied, too, only perhaps later and > with a different frequency pattern.
+1 can confirm - I also see the stack trace in question [1] from time to time, however even under low/moderate load and more or less randomly. This also used to happen with 3.14.x, so I don't think it has anything to do woth the BQL patch pe se.
I *think* - and have suspected for some time now - that this is caused by a change in TSO/GSO or other offload settings, and that some r8169 models consequently suffer occasional (or in your case frequent) hiccups. BQL and/or a qdisc doing requeues might just interfer with whatever upsets the driver or HW.
I used to turn on all possible settings via ethtool, but am now back to defaults, which disables sg/tso/gso. Exactly which of these settings is the culprit remains to be seen.
Holger
[1] https://lkml.org/lkml/2014/12/11/65
| |