openSUSE / SUSEPrime

Provide nvidia-prime like package for openSUSE
64 stars 11 forks source link

Revert Blacklist ipmi_msghandler, ipmi_devintf kernel modules #71

Closed HollowMan6 closed 2 years ago

HollowMan6 commented 2 years ago

Unloading ipmi_msghandler and ipmi_devintf will stop my RTL8821CE WiFi card from working, and I also agree with what randombk said here, IPMI may not be the root cause for unable to turn off NVIDIA GPU.

As a result, although the problem doesn't occur in my system, I suggest looking for other true culprits for the situation described here.

And according to what thejinx0r suggested, I made a patch to stop and disable nvidia-persistenced service before unloading Nvidia drivers. Hope it helps.

Other possibilities may include onsc's situation, but I guess we can do little for it in SUSE Prime select.

sndirsch commented 2 years ago

Well, this will break again https://github.com/openSUSE/SUSEPrime/issues/50#issuecomment-605084071 For sure I wasn't using nvidia-persistenced when testing.

HollowMan6 commented 2 years ago

I guess it may need further discussion for https://github.com/openSUSE/SUSEPrime/issues/50#issuecomment-605084071 . I'm sure blacklisting ipmi_msghandler, ipmi_devintf will stop my RTL8821CE WiFi card from working, and I guess such a situation will also happen when someone uses other similar hardware and that shouldn't be the solution imo.

Sorry I can't help further since I can't reproduce the situation described in https://github.com/openSUSE/SUSEPrime/issues/50#issuecomment-605084071 , I can successfully turn off my NVIDIA card when using intel without blacklisting ipmi_msghandler and ipmi_devintf, not sure why that would happen in that case.

Maybe you can get more inspiration from other comments in this issue? https://github.com/Bumblebee-Project/bbswitch/issues/173

HollowMan6 commented 2 years ago

According to IBM's documentation, ipmi_* drivers correspond to Intelligent Platform Management Interface (IPMI) (Wikipedia). As it provides the ablities to monitor hardware, does the situation here https://github.com/openSUSE/SUSEPrime/issues/50#issuecomment-605084071 means that some software are still monitoring or using NVIDIA card when SUSE prime select unloads the NVIDIA driver?

So the utimate solution maybe kill all the nvidia related process before unloading, just like here https://github.com/Bumblebee-Project/bbswitch/issues/173#issuecomment-439660939 ? Can you try it since I can't reproduce the situation @sndirsch .

sndirsch commented 2 years ago

Thanks @HollowMan6 I'll try to test this again next week ...

HollowMan6 commented 2 years ago

Adapt according to https://github.com/Bumblebee-Project/bbswitch/issues/173#issuecomment-439660939 , hope that helps.

sndirsch commented 2 years ago

Ok. Tried to reproduce the issue with loaded ipmi_* kernel modules with current Tumbleweed and nvidia G05 driver. They are (no longer?) loaded and needed by nvidia kernel module. Hope I tested on the same machine at that time ... so I'm going to accept this pull request. :-)