Cannot change power limits for Ryzen 7 8840U

[KOPRajs]

I'm running RyzenAdj in Ubuntu 24.04 on a HP laptop with Ryzen 7 8840U. I can change the STAPM temp and I am also able to lower the TDP values, but I can't raise the values above the default.

Currently stuck at 25 W for continuous load, even though the APU temp is kept below 70C. I'm not using secure boot and I have tried using both the ryzen_smu kernel module and the direct /dev/mem approach, but the results are the same.

Hardware:
HP EliteBook 845 G11
AMD Ryzen 7 8840U

Default values:
STAPM LIMIT - 30.000
PPT LIMIT FAST - 30.000
PPT LIMIT SLOW - 25.000

I expect that it is being somehow limited by AMD or HP, but is there a chance RyzenAdj can overcome this in the future? If the limit cannot be changed, would it be possible to fool the APU to run on the FAST limit for a longer time? Are there any other methods worth trying (e.g. https://github.com/sibradzic/stapmlifier) or will it be limited the same as the RyzenAdj?

My goal would be to raise the limit to at least 30 W, even 35 if it is possible, because currently the GPU is limited to cca 1700 MHz in continuous loads like games.

[KOPRajs]

I've just found the explanation here: https://github.com/FlyGoat/RyzenAdj/wiki/Supported-Models#custom-value-range-for-zen3

I'll try the suggested workarounds and will report back.

[KOPRajs]

It turned out that I can't change the STAPM limit at all (not even to lower values), so the workaround described in the above link cannot be used.

I will try to get the stapmlifier method to work when I get some time. Unfortunately, I'm on Ubuntu stock kernel, which is missing the custom_method kernel module, so even if it works, it is going to be a pain maintaining it.

[KOPRajs]

It seems almost impossible to build the custom_method kernel module for recent distribution kernels to use the stapmlifier as intended anymore, see sibradzic/stapmlifier#7.

However, by looking at the code, it just constructs a buffer with parameters and call the \_SB.ALIB ACPI function. It might be possible to do it manually without the need for stapmlifier and the custom_method kernel module.

I've tried something like this:
echo -e "\_SB.ALIB 0x0C $(printf '\\x07\\x00\\x06\\x30\\x75\\x00\\x00')" | sudo tee --append /proc/acpi/call

The hex values contain the parameters:
x07 - the size of the buffer (constant)
x06 - DPTCi code
x30 x75 - little-endian representation of 30000

The return value in /proc/acpi/call is 0, but it seems to have no impact on the PPT LIMIT SLOW (and FAST in this example) so far.

Any ideas are welcome.

[KOPRajs]

While I was not able to get the deprecated custom_method module to work with the stock Ubuntu kernel, I've been able to modify the stapmlifier source code (only the header), so it can be used in the SSDT table instead of the DSDT and I have successfully loaded it using the acpi_configfs kernel module.

This command now works for me:
echo "\STPM 30000 0x07" | sudo tee --append /proc/acpi/call

However, there seems to be the same limitation as when using RyzenAdj directly, I can change the limit below the default, but I can not increase the limit above its default value (I'm stuck with 25001 mW).

I don't know, if the limitation comes from using SSDT table instead of the DSDT, or if it wouldn't work via the deprecated custom_method module either.

EDIT: The last method I'm going to try is changing the limits directly in the DSDT, but currently I can't recompile my DSDT table even without any modifications, see here: acpica/acpica#1049