GPU supported RTD3 block CPU PC10 after `prime-select intel`

[alex-tu-cc]

From Nvidia 450 a group of GPU support RTD3(by referring supported-gpus.json) can reach CPU PC10 in on-demand mode (prime-select on-demand). But failed on power-saving mode(prime-select intel)

reproduce steps:

1. $ prime-select on-demand
2. get system to long idle
3. SSH$ sudo turbostat --show GFX%rc6,Pkg%pc2,Pkg%pc3,Pkg%pc6,Pkg%pc7,Pkg%pc8,Pkg%pc9,Pk%pc10,SYS%LPI GFX%rc6 Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 Pkg%pc8 Pkg%pc9 Pk%pc10 SYS%LPI 100.67 5.16 0.87 0.01 3.41 0.77 2.81 85.86 0.00
4. $ prime-select intel
5. get system to long idle
6. SSH$ sudo turbostat --show GFX%rc6,Pkg%pc2,Pkg%pc3,Pkg%pc6,Pkg%pc7,Pkg%pc8,Pkg%pc9,Pk%pc10,SYS%LPI GFX%rc6 Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 Pkg%pc8 Pkg%pc9 Pk%pc10 SYS%LPI 100.03 1.85 97.39 0.00 0.00 0.00 0.00 0.00 0.00

Note: this issue not related to https://github.com/tseliot/nvidia-prime/pull/12, the issue can be reproduced without that patch.

The tested Nvidia gpu:

01:00.0 3D controller [0302]: NVIDIA Corporation TU117GLM [Quadro T2000 Mobile / Max-Q] [10de:1fb8] (rev a1)
        Subsystem: Dell TU117GLM [Quadro T2000 Mobile / Max-Q] [1028:097e]
        Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx+
        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-

Comparing to another Nvidia GPU which NOT supported RTD3, it can get CPU pc10 well in power-saving mode (prime-select intel)

another Nvidia GPU

01:00.0 3D controller [0302]: NVIDIA Corporation GP108M [GeForce MX330] [10de:1d16] (rev a1)
        Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
        Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-

[alex-tu-cc]

$ /var/log/gpu-manager.log | pastebinit https://paste.ubuntu.com/p/fXsN6f4wFD/

[alex-tu-cc]

I can not reproduce this issue anymore. I'm wondering is it caused by the same root cause from https://github.com/tseliot/nvidia-prime/issues/14 which seems the modprobe configurations in initrd is in-sync from runtime rootfs.