Bug#843349: linux-image-4.7.0-1-amd64: Intermittently high system load, sluggish response, no /proc/<pid>/stat utime,stime,etc. reported

Just an update: I am still encountering similar performance issues with the 4.9.0-1 kernel that was recently promoted to testing. In fact, the 4.9.0-1 kernel may be the worst yet in terms of this specific issue, with IntelliJ starting so slowly as to be effectively unusable. One time the entire system appeared to become completely unresponsive & a hard reset was required.

The most recent kernel with which I've experienced problems:

$ apt-cache show linux-image-4.9.0-1-amd64

Package: linux-image-4.9.0-1-amd64

Source: linux-signed (4)

Version: 4.9.2-2

Installed-Size: 183299

This problem does not seem to occur with a vanilla 4.9.2 kernel, implying something Debian-specific:

$ uname -a

Linux desktop.local 4.9.2 #1 SMP Wed Jan 25 04:28:29 PST 2017 x86_64 GNU/Linux

If you're aware of any Debian patches that might be suspect, I'm happy to try some custom builds. Reproducing the issue is trivial on my end when a kernel is "bad".

On Sun, Nov 20, 2016 at 2:24 PM, Tom Lee <debian@tomlee.co> wrote:

Also potentially relevant: kworker/0:0 has 421h on-cpu time reported per htop (the TIME+ column) and top. /proc/<pid>/stat output for PID=4 (kworker/0:0) vs PID=63 (kworker/1:1):

tom@desktop ~ $ cat /proc/63/stat
63 (kworker/1:1) S 2 0 0 0 -1 69238880 0 0 0 0 0 0 0 0 20 0 1 0 82 0 0 18446744073709551615 0 0 0 0 0 0 0 2147483647 0 0 0 0 17 1 0 0 184 0 0 0 0 0 0 0 0 0 0
tom@desktop ~ $ cat /proc/4/stat
4 (kworker/0:0) S 2 0 0 0 -1 69238880 0 0 0 0 151836336 0 0 0 20 0 1 0 10 0 0 18446744073709551615 0 0 0 0 0 0 0 2147483647 0 0 0 0 17 0 0 0 0 0 0 0 0 0 0 0 0 0 0

In fact, kworker:0:0 seems to be the only process with on-cpu stats reported (forgive the shell abuse, quick and dirty):

$ for pid in $(ls /proc | awk '/^[0-9]*$/'); do [ -d /proc/$pid ] && awk '$14 != "0" { print $2 }' /proc/$pid/stat; done
(kworker/0:0)
$

On Sun, Nov 20, 2016 at 2:12 PM, Tom Lee <debian@tomlee.co> wrote:
Tried 4.8.0-1 with similar results. Perhaps slightly better, but not much.

$ uname -a
Linux desktop 4.8.0-1-amd64 #1 SMP Debian 4.8.5-1 (2016-10-28) x86_64 GNU/Linux

Like 4.7, can't see what's hogging the CPUs as per-process CPU stats don't appear to be correctly reported in /proc/<pid>/stat.

4.6.x continues to be perfect with respect to both overall performance and the correctness of /proc/<pid>/stat.

On Sat, Nov 12, 2016 at 8:23 PM, Ben Hutchings <ben@decadent.org.uk> wrote:
Control: tag -1 moreinfo

On Sun, 2016-11-06 at 00:02 -0700, Tom Lee wrote:
[...]
> Downgrading to linux-image-4.6.0-1-amd64 fixed all symptoms. Haven't yet
> tried 4.8.0-1 from unstable, not sure if it's impacted.
[...]

Please do.

Ben.

--
Ben Hutchings
Nothing is ever a complete failure; it can always serve as a bad
example.

--
Tom Lee / http://tomlee.co / @tglee

--
Tom Lee / http://tomlee.co / @tglee

Tom Lee / http://tomlee.co / @tglee