BalancedDef and PowersaverDef set to -1 in all configs effectively disables lpmd when power-profiles-daemon is running

[ptr1337]

All shipped configuration files set -1 and -1, which forces lpmd OFF for both the "balanced" and "power-saver" power profiles. Combined with -1, this means lpmd is unconditionally disabled whenever power-profiles-daemon
(PPD) is running.

Since PPD is active by default on most modern Linux desktop environments (GNOME, KDE), and the default profile is "balanced", lpmd will never engage on a typical installation.

The daemon only becomes active when PPD is not installed, because connect_to_power_profile_daemon() failure causes a fallback to LPMD_AUTO in lpmd_proc.c.

  ┌──────────────────────────────────────────────┬─────────────┬───────────────┐
  │                 Config file                  │ BalancedDef │ PowersaverDef │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config.xml                        │ -1          │ -1            │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config_examples.xml               │ -1          │ -1            │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config_experimental.xml           │ -1          │ -1            │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config_F6_M170.xml (Meteor Lake)  │ -1          │ -1            │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config_F6_M189.xml (Lunar Lake)   │ -1          │ -1            │
  ├──────────────────────────────────────────────┼─────────────┼───────────────┤
  │ intel_lpmd_config_F6_M204.xml (Panther Lake) │ -1          │ -1            │
  └──────────────────────────────────────────────┴─────────────┴───────────────┘

Expected behavior

For mobile/laptop platforms where lpmd is designed to save power, more useful defaults would be:

• PerformanceDef=-1 — keep lpmd off when user explicitly wants maximum performance (correct as-is)
• BalancedDef=0 — allow lpmd to opportunistically enter/exit low power mode based on WLT hints and utilization
• PowersaverDef=0 — same as balanced, or potentially 1 (force on) to aggressively save power

This way lpmd actually participates in power management on systems where PPD is running, which is the common case on desktop Linux distributions.

[ptr1337]

Why is this even done? Does power-profiles-daemon conflict with the profiles?

I think this could be much improved :)

[spandruvada]

When you turn on by default, this creates issues with benchmarking. When daemon is loaded there will be some latency to switch to full power mode, so some benchmarks are sensitive. Also some benchmarks looks at the available CPUs at start and then start using lower CPUs always.

So this is up to distro what is the use case.

Users can always enable either by config file
or do "intel_lpmd_control AUTO"

[ptr1337]

When you turn on by default, this creates issues with benchmarking. When daemon is loaded there will be some latency to switch to full power mode, so some benchmarks are sensitive. Also some benchmarks looks at the available CPUs at start and then start using lower CPUs always.

So this is up to distro what is the use case.

Users can always enable either by config file or do "intel_lpmd_control AUTO"

Not sure about this, because this makes whole lpmd integration meaningless for most users and most distributions are not implementing lpmd yet.

Lot of people (see issues here) are confused why lpmd is not properly working often.

[maciejwieczorretman]

@spandruvada maybe we could add a flag for benchmarking? It could override the config file and initialize modes to -1 for benchmark purposes.
And the defaults could be more user friendly as @ptr1337 suggested:
PerformanceDef=-1
BalancedDef=0
PowersaverDef=1

[spandruvada]

Some part of LPMD doesn't enable. For example slider setting. They are applied in Panther Lake going forward.
The only part about CPUs is tough as benchmarks runs with fewer CPUs. Need to think how benchmarking can be identified to remove CPU isolation part.

[ptr1337]

Actually, ive been looking forward implementing this as default into CachyOS, see CachyOS/chwd#233 to provide on mobile platforms a better experience. Lunar and Panther Lake are already incredible good in terms of battery life, but the scheduling needs to improve and LPMD helps there a bit.

Ive forked for now lpmd and we will provide use some different profiles, see https://github.com/CachyOS/intel-lpmd/commits/cachyos . Right now ive set the value to Auto, but likely going to move it for powersave to 1.

But I think it would be great, if this could be handled directly via upstream in the near future to improve this - or consider reattempting to work on the Thread Director for Linux - this one got stalled: https://lore.kernel.org/lkml/20230613042422.5344-1-ricardo.neri-calderon@linux.intel.com/

[maciejwieczorretman]

I'm resolving one issue right now but after that I wanted to add the ability to pass config file path as a parameter, would that be helpful for this case?

[ptr1337]

I'm resolving one issue right now but after that I wanted to add the ability to pass config file path as a parameter, would that be helpful for this case?

Sure, that would help. Then distribution/users can use the provided profiles as reference and then improve by the needs and pass to an own config.