Offline upgrades and Power Management

Project Discussion
, ,
1

Ryzen-based laptop is getting hot and noticeably consuming battery during some offline updates.

There are quite a few topics like Fedora burns twice as much power as Ubuntu and articles like Making sure you're not a bot! dedicated to power management tuning but they don’t cover this aspect.

Preset TuneD profiles (like powersave) do tweak CPU Boost/EPP, but it is not propagated to the update environment through bootloader plugin or other means.

Quick and dirty solution I see so far is custom laptop-specific dracut module with suitable hardcoded defaults and complementing custom laptop-specific TuneD-ppd profiles.

Has this been discussed/considered previously? Is this the desired behavior? Does generic enough dracut/tuned/bootloader/cmdline/initrd solution looks like overkill for regular boot sequences?

2

After more digging and testing, it appears that the system fully supports persistent ACPI platfrom_profile configuration, but TuneD messes it up.

First of all, it defaults to auto value of main.rollback instead of not_on_exit, effectively and intentionally preventing persistence of the tuneable between reboots.

But switching configuration to not_on_exit does not help, because in this mode TuneD does not rollback if I stop tuned service through systemctl stop but it does rollback when exiting due to system shutdown.

Having expected platfrom_profile value noticeably improves moderate+sized offline upgrade experience for this laptop. Not as good as having quick and dirty solution for EPP and Boost in addition, but much better than nothing.

Is the decribed TuneD behavior intentional?

Is it reasonable to alter TuneD rollback logic to skip rollback on shutdown when main.rollback=not_on_exit and have it as part of default TuneD configuration for Workstation/Desktop editions, or complete exclusion of /sys/firmware/acpi/platform_profile interface from TuneD rollback logic?

4

I’m not sure anyone with deep enough knowledge to confirm or deny your diagnosis is reading here. If you think you’ve spotted a specific bug in tuned, the best thing to do would be to file a report.

1 Like
5

I am not sure. Judging by the source code[1][2] the distinction is kinda intentional, so answer to is it TuneD bug or not probably depends on the current vision, if there is any, on how tuned, tuned-ppd and DE profile selection should influence (or not) various boot stages and sequences on various hardware. akmods is another case that comes to mind, but I haven’t looked at it yet.

I found Power Management SIG and relevent mail list but these appear inactive, so I can’t figure out who is making decisions here within Fedora Project.

Should I just create a specific tuned ticket in Fedora Bugzilla about main.rollback=not_on_exit behavior and dump the findings so far there?

6

Sure, I guess that should work. AFAIK tuned is a Red Hat / Fedora joint, so whether you file up or downstream you’ll probably hit the same people.

7

Thanks! Posting link in case someone here is interested 2537319 – Can't prevent ACPI power_platform rollback on system shutdown using rollback=not_on_exit configuration