[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Desktop stutter with 5.10



Hello,

I'm seeing some performance issues on my Dell laptop (stock Debian
bullseye, XFCE 4.16, kernel 5.10) that show up on boot and vanish after
a suspend-resume cycle:

- before suspending to RAM:
    - the desktop stutters: visible typing latency, application startup
      delay, PDF scrolling glitches;
    - idle frequency[1] stays around 1.70 GHz.

- after resuming from suspension:
    - I observe no stutter anymore;
    - idle frequency[1] stays around 500-700 MHz.

In both cases, /sys/devices/system/cpu/cpuferq/policy*/scaling_governor
say "schedutil".


See [2] for more details about my setup.  I started observing this on
Debian buster when upgrading to backported kernels: performance was fine
up to kernel 5.7; then I started noticing these hiccups with kernel 5.8.

I asked for advice then[3], when 5.9 was released[4], and when 5.10
found its way in buster-backports I tried to cook up a somewhat
quantifiable recipe showcasing the issue[5].

I had hoped to chalk these problems up to some backport mishap;
unfortunately, the delay and variance in application startup reported in
[5] are still representative of what I observe on a fresh bullseye
install:

- before suspending to RAM, application startup times vary wildly,
  averaging on 2x,
- after resuming from suspension, application startup times drop to x.

(Same goes with other, less quantifiable issues, e.g. typing latency,
PDF scrolling glitches… which go away after a suspend-resume cycle.)


What would be my best hope for getting to the bottom of this problem?
Is there enough information here for a bug report?  Should I look for
help on another list?

FWIW, the Internet is chock full of people who experienced issues with
5.8's changes wrt intel_pstate[6], but I'm not sure these are the same
issues: most of them describe a CPU stuck at low frequencies, which is
not my problem: my idle frequency drops *after* resuming, and does *not*
negatively impact performance, oddly enough.

FWIW², I added intel_pstate=active to the kernel command-line, checked
that /sys/devices/system/cpu/intel_pstate/status reported "active" (it
says "passive" ordinarily), and AFAICT the symptoms persisted.


Thanks for your time.


[1] As eyeballed with 'watch cpupower -c all frequency-info'.

[2] - Dell Latitude 3190 (4× Intel® Celeron® N4100)
    - Fresh install of stock Debian 11 / XFCE 4.16 / kernel 5.10
    - Packages from non-free component:
      firmware-iwlwifi, firmware-misc-nonfree

[3] https://lists.debian.org/debian-kernel/2020/11/msg00084.html
[4] https://lists.debian.org/debian-kernel/2020/12/msg00040.html
[5] https://lists.debian.org/debian-kernel/2021/02/msg00231.html

[6] E.g.
https://bugzilla.kernel.org/show_bug.cgi?id=90041
https://bbs.archlinux.org/viewtopic.php?id=264080
https://bbs.archlinux.org/viewtopic.php?id=260361
https://forum.manjaro.org/t/kernel-5-8-x-my-system-seems-to-be-only-running-at-it-s-minimum-clock-of-800mhz/2919
https://unix.stackexchange.com/questions/625563/cpufreq-scaling-broken-on-i7-1185g7-kernel-5-8-0
https://old.reddit.com/r/debian/comments/5lt2e8/dell_intel_pstate_and_a_cpu_stuck_at_minimum/
https://old.reddit.com/r/Dell/comments/5uh6wo/fixing_your_dell_xps_cpu_stuck_at_minimum/
https://old.reddit.com/r/linux/comments/ihdozd/linux_kernel_58_defaults_to_passive_mode_for/


Reply to: