[question] CPU core unpinning?

[stereomato]

Basically, I've had an issue with intel_lpmd, which is that if I have to disable it (either by stopping the service or switching to performance mode on gnome's system menu) if I want to do anything I know is demanding, like a game, because otherwise I'll get awful performance in such situations, as the software in question only seemingly is allowed to use the last 4 threads, as shown by htop (13-16). So, is there a way for intel_lpmd to unpin stuff from the low power cores (i'm on alderlake i5-12500h)

And yes, I run intel_lpmd always on AUTO mode, regardless of being on AC or BAT power. I'm on Arch Linux as well.

[zhang-rui]

Basically, I've had an issue with intel_lpmd, which is that if I have to disable it (either by stopping the service or switching to performance mode on gnome's system menu)

Switching to performance mode means you set lpmd to Forced OFF instead of AUTO.

if I want to do anything I know is demanding, like a game, because otherwise I'll get awful performance in such situations,

IMV, performance mode would be a better choice of scenarios like games.

as the software in question only seemingly is allowed to use the last 4 threads, as shown by htop (13-16).

can you attach the config file you're using?
It seems that lpmd is configured improperly. Or else, lpmd should active all CPUs when CPU 13-16 utilization is high.

[stereomato]

IMV, performance mode would be a better choice of scenarios like games.

While yeah, the balanced power profile gives me the performance I need, and it doesn't require me to remember to switch to it.

can you attach the config file you're using?

It seems that lpmd is configured improperly. Or else, lpmd should active all CPUs when CPU 13-16 utilization is high.

I've set 20 and 25 on the enter/exit parameters so that the system can enter low power more often, but also exit it quickly. With the default settings it behaves the same anyway.

intel_lpmd_config.txt

[zhang-rui]

I don't think the problem should exist in any release of the lpmd, there must be something wrong. But still, I'd like to confirm that, does the problem still exist in the latest release or git?
If yes, please run the latest version with --loglevel=debug, and attach the log here.

[stereomato]

Log included. I compiled the git version thru the intel-lpmd-git aur package, the issue happens indeed here. If I set the system to performance mode (which disables intel_lpmd) the games runs normally, and even loads faster.

intel_lpmd_debug.log

[zhang-rui]

From the log, what I can see is that the system enters low power mode for a short period and then quits immediately. And this pattern repeats for a couple of times.
I think the problem here is that the current threshold and sample period setting is not suitable for this workload, thus the system keeps on switching between full_power and lpm_deep.

Does the workload type hint work on your systems? can you try the wlt monitor instead of utilization monitor?

[stereomato]

I use alder lake, how do I enable the workload type hints?

[stereomato]

enabling WLTProxyHint and WLTHintEnable was enough for that. I did also run intel_lpmd_control AUTO before (when I only enabled WLTHintEnable) because just enabling that didn't seem to do anything, but after I enabled WLTProxyHint it started just working. Can't test a heavy workoad right now, though

[stereomato]

I think WLT should be added to the xml config file by default. Seems to work well on my alder lake laptop, and given that it's a cheap asus laptop, I think it would work just fine on any other laptop or computer.