Hello,
This is a followup to my previous thread[1] where I observed some
stutter after upgrading my kernel from 5.7 to 5.8 on Debian buster,
using the backports repository.
Since 5.10 is now available, I have tried this version out, and I still
observe the issue. I have found a workaround, however: for some reason,
the stutter goes away after a suspend-resume cycle.
In my original thread, I recounted a few ways this stutter manifests
(typing latency, application startup delay, PDF scrolling glitches); I
have since found a quantifiable reproducer:
for ((i=0; i<100; i++)); do
time -p terminator -x 'bash -c exit'
done &> time-$(uname -r)
grep $mode time-$(uname -r) | sort -k2
Where $mode is either real, user or sys. Here are the extrema:
real user sys
min max min max min max
5.7 0.44 0.48 0.29 0.35 0.02 0.08
5.10 0.47 3.30 0.31 2.87 0.02 0.30
5.10-after-suspend 0.46 0.50 0.30 0.36 0.03 0.08
Averages & standard deviations:
real user sys
5.7 0.454±0.006 0.322±0.013 0.046±0.013
5.10 0.961±0.607 0.559±0.518 0.074±0.052
5.10-after-suspend 0.469±0.005 0.332±0.012 0.050±0.011
These results feel consistent with the stutter I observe: 5.7 and
5.10-after-suspend have roughly the same performance, while
freshly-booted 5.10 is overall very jittery: terminator startup is twice
as slow on average, with 100× more variance.
Some more stats, eyeballed from /proc/cpuinfo and
/sys/devices/system/cpu/cpufreq/policy*/scaling_governor:
idle frequency CPU governor
5.7 < 800 MHz powersave
5.10 > 1900 MHz ondemand
5.10 (after suspend) < 900 MHz ondemand
I've attached the output of dmesg on 5.7 and 5.10 (which includes the
suspend-resume cycle). I couldn't make heads or tails of it, but maybe
it holds a clue?
Has this become precise enough to warrant a bug report?
Thank your for your time.
[1] <87tutrlhle.fsf@gmail.com>
Attachment:
dmesg-5.7.0-0.bpo.2-amd64.xz
Description: application/xz
Attachment:
dmesg-5.10.0-0.bpo.3-amd64-after-suspend.xz
Description: application/xz