Resume from sleep problem (with Nvidia, Xorg, KDE Plasma)

[mcnutty]

Resuming from sleep often results in a broken/unusable system. I've been using an nvidia card with nvidia-drivers, xorg and KDE/Plasma with multiple monitors for a long time. The exact card, kernel, and software versions have varied, but I have had this problem for many years. My previous cards/systems didn't use all that much power when idling so I basically just gave up trying to solve the issue. However, I recently got a new graphics card that uses significantly more power at idle so I thought I'd try again, but I still run into the problem and it prevents me from being able to put the system to sleep, which is now a larger concern.

I would say about 50-75% of the time the system wakes up and everything is fine. However, the other 25% there is a significant problem that requires a reboot. Sometimes the system just doesn't ever really wake up. The computer appears to wake up, but the monitors never turn on and the system basically just hangs. Other times the whole system wakes up and all the monitors turn on. I'm able to unlock the screens and get back in, but the main panel is gone and other parts of the GUI don't seem to work properly, for example using the scroll wheel on a blank area does not switch virtual desktops like usual.

I've seen threads like these, but none seem to point to a solution that works for me.

https://forums.gentoo.org/viewtopic-t-1155841-highlight-nvidia+suspend.html
https://forums.gentoo.org/viewtopic-t-1133268-highlight-.html
https://bugs.gentoo.org/693384
https://bugs.kde.org/show_bug.cgi?id=356727

Any ideas on where the root problem is, where to look for solutions, or where to file a bug report?

[pietinger]

mcnutty,

with all what I know about kernel and nvidia I can give only these suggestions:

a) Dont use nvidia cards, or
b) Dont use Suspend, or
c) Wait for new kernel and nvida driver versions.

Sorry, no joking here

[mcnutty]

Yeah, I was afraid of that, but not exactly unexpected. Unfortunately, I need the card for machine learning so nvidia is basically a requirement. I'll continue to stick with (b) for now and hope that pigs will fly someday.

[logrusx]

This is a very generic question.

Please provide details for your system.

For the short while I used the dGPU on my laptop it worked quite fine with 5.15 kernel and 510 driver versions. One thing is it needed the nvidia-suspend/resume/hibernate services turned on to do some stuff before sleep and after wake up.

I believe systemd is useful here to trigger all that, I don't know how openrc performs.

Best Regards,
Georgi

[rab0171610]

I have used KDE practically since its inception. I also have always used Nvidia cards and proprietary drivers. I have used every distro imaginable. I have never found suspend/resume/hibernate to be reliable. There has always been some issue or other. It seemed like I was always having to reboot just to get things back after sleep. These issues were not always limited to Linux but could also appear in Windows. I gave up on it years ago. I resorted to power settings that will turn the monitor off after a period of inactivity and if I leave the system for more than a few hours I just shut the system down. I have grown accustomed to it. In my own experience, sleep has always been more like a bug than a feature. I don't think it is an essential feature to say the least.

[Hu]

Although this will not help OP since the target system requires the proprietary nVidia drivers, I will note that when last I used a system with an nVidia card, I used the Nouveau driver for it and had no problems with suspend or with hibernate (although, the standard disclaimer applies that the open drivers do not drive the card to its full potential, due to missing documentation from nVidia). nVidia hardware can be made to do this right. The problem is that apparently their proprietary drivers get it wrong.

[mcnutty]

Thanks to everyone for the responses and moral support if nothing else.

[logrusx]

[mcnutty]

I'm not sure I'm ready to declare complete victory and mark this as solved yet, but upgrading to KDE/Plasma 6 has at least significantly reduced the problem. After about 10-20 sleep/wake cycles, I have yet to hang.

[shazeal]

[Ionen]

ftr https://github.com/gentoo/gentoo/pull/38482 is coming soon, not to say it will necessarily solve all issues and work for everyone (I know it doesn't), but a few users with several combinations of nvidia+systemd/elogind+wayland/xorg+plasma6 have successfully used suspend and resumed back into a system that seems to work as intended, so it should at least not be entirely broken by default.

PR could potentially make things worse for setups that don't use elogind nor systemd to sleep though (but easy to set back to =0 if needed), this is primarily catering to what Plasma and Gnome wants.

[Child_of_Sun_24]

I have Plasma6 running on Wayland with nvidia-drivers-550.107.02-r1 and i am using systemd as init system.

This is my nvidia.conf:
http://0x0.st/Xxnm.conf

Important is the use of "NVreg_PreserveVideoMemoryAllocations=1" and i have to enable the "nvidia-hibernate.service nvidia-resume.service nvidia-suspend.service", with this configuration suspend and resume are no problem, even with older drivers.

Under openrc i have seen a shell script which imitates the behavior of "nvidia-hibernate.service nvidia-resume.service nvidia-suspend.service" it can be found here https://forums.gentoo.org/viewtopic-p-8813895.html?sid=05052e14289f0e63a8413e0cc21f51cb but i haven't testet it because i am using systemd.

I hope this helps a bit.

[Ionen]

The PR in the post above yours essentially does that, it enables these services by default, installs a elogind hook, and sets =1.

It's also merged now, aka nvidia-drivers-550.107.02-r1 has these changes. Albeit if you already had =1, use systemd, and enabled the services, it does nothing new for you

[eeckwrk99]

My setup:

- GTX 970
- systemd
- X11
- AwesomeWM

Now that NVreg_PreserveVideoMemoryAllocations=1 is the default, I had to revert the change as it completely breaks suspend/hibernate on my end:

[logrusx]