smsc95xx 1-1.1:1.0 eth0: kevent 0 may have been dropped

[andersthomson]

On a recent install of rpi-4.9.y, 4.9.28, I get this after a period of constant tv streaming.

smsc95xx_eth0_kevent_0_may_have_been_dropped.txt

Didn't get it all, so I attach frech dmesg as well to get a view on the early parts.

4.9.28_dmesg.txt

[JamesH65]

Interesting. The kevent dropped message occurs when the system is trying to deal with a particular scenario using kevents (using the system queue), then the scenario occurs again before the previous one has been handled. However, to get that many in the log implies that the entire processing of those messages has died completely. Usually just one every now and again. Is this repeatable? Did the TV streaming die?

[andersthomson]

Yep it died. Probaby as en effect, not cause, of this. No network connectivity after this event. I've seen it on vanilla 4.11 and, iirc, rpi-4.9.y 4.9.27 as well.

…

On May 19, 2017 4:21:40 PM GMT+02:00, James Hughes ***@***.***> wrote: Interesting. The kevent dropped message occurs when the system is trying to deal with a particular scenario using kevents (using the system queue), then the scenario occurs again before the previous one has been handled. However, to get that many in the log implies that the entire processing of those messages has died completely. Usually just one every now and again. Is this repeatable? Did the TV streaming die? -- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: #2027 (comment)

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

[JamesH65]

yes, if you get that kevent dropped message, the network stack is probably FUBARd.

The log seems to imply bridging is set up - can you describe how the device is set up please?

[popcornmix]

Does increasing vm.min_free_kbytes in etc/sysctl.d/98-rpi.conf help?
e.g. double it to 32678.

[andersthomson]

It's running 4 systemd-nspawn containers via its bridging support. I'll try the min_free_kbytes when I get back to the device end of the week. This one is fairly easily triggered. I've seen it about 6 times already this week on both .26 and .27 kernels. .17 does not have this issue, so my plan was to move upwards from that to see what triggered it.

[JamesH65]

What are you running in the containers? I wonder if the CPU load of running all the containers starves the main system kevent queue of time, so the kevent required to process the ethernet packet problem isn't run in time, or whether the system queue is simply swamped with events that need processing.

[andersthomson]

One containter with tvheadend, one with openhab, one with nginx, one for misc cli stuff. In terms of load, tvheadend does the most net traffic. nginx is virtually dormant, so is openhab.

[andersthomson]

Hi!
Any update on this? (What does 'Assigned for implementation' mean?)

I just updated to rpi-4.9.33 and have had two instances like this in like 24h or so.
I previously ran a patched (the cgroup thing) 4.9.30 which ran fine for a couple of weeks.

I'd be happy to test out any patches you have...

[JamesH65]

There was a fix for an issue that caused this when bridging a month back, so it's possible you already have it. There may be further error paths that cause it though. Did you provide a means of replication?

…

On 20 Jun 2017 18:32, "andersthomson" ***@***.***> wrote: Hi! Any update on this? (What does 'Assigned for implementation' mean?) I just updated to rpi-4.9.33 and have had two instances like this in like 24h or so. I previously ran a patched (the cgroup thing) 4.9.30 which ran fine for a couple of weeks. I'd be happy to test out any patches you have... — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#2027 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADqrHWC9twUQRFIQFbwghVhenrULoq13ks5sGAIkgaJpZM4NgiiZ> .

[andersthomson]

If it's in 4.9.33 I should have it. Unfortunately, I have no means to reproduce it. No idea what triggers it (last one seems to have started at 18:00 sharp, fwiw). Any chance I can sprinkle a few printfs in selected places to get more context out?