My system:
- EndeavourOS Linux x86_64
- XPS 15 9500
- 5.17.5-arch1-1 kernel
- NVIDIA GeForce GTX 1650 Ti Mobile with nvidia-dkms 510.68.02-1 driver
Description of the problem:
I have an Optimus laptop with an NVIDIA Turing card and I have already configured the RTD3 power management following the Arch Wiki. I’ve also configured the system to preserve video memory after suspend and enabled nvidia-suspend, nvidia-resume, nvidia-hibernate and nvidia-persistenced service.
When I put the laptop into either suspend or hibernate state, using either systemctl suspend or systemctl hibernate, everything works as expected. However, if I use systemctl suspend-then-hibernate the system fails to enter in suspend state with the following error message:
mag 07 15:09:00 ikigai systemd[1]: Starting Suspend; Hibernate if not used for a period of time...
mag 07 15:09:00 ikigai systemd-sleep[292811]: Entering sleep state 'suspend'...
mag 07 15:09:07 ikigai systemd-sleep[292811]: Failed to put system to sleep. System resumed again: Input/output error
mag 07 15:09:09 ikigai systemd[1]: systemd-suspend-then-hibernate.service: Main process exited, code=exited, status=1/FAILURE
mag 07 15:09:09 ikigai systemd[1]: systemd-suspend-then-hibernate.service: Failed with result 'exit-code'.
mag 07 15:09:09 ikigai systemd[1]: Failed to start Suspend; Hibernate if not used for a period of time.
My guess is that there is no nvidia-suspend-then-hibernate service. Specifying NVreg_PreserveVideoMemoryAllocations=0 works, but only if the NVIDIA card is not in use: if it is, then the system sometimes goes to sleep and sometimes not, causing unexpected overheat (e.g. when I close the lid and put the laptop is in my bag).
Unfortunately my laptop does not support s3 (“deep standby”) but only s2idle (“modern standby”): thus when suspended the system still consumes a lot of power (30% in 10 hours) and for this reason I need to use the suspend-then-hibernate feature.
Am I missing something? Does someone else have this problem?
1 Like
I think I’ve found a workaround: I disabled nvidia-suspend, nvidia-hibernate and nvidia-resume and I’ve edited the /lib/systemd/system-sleep/nvidia as follows:
#!/bin/sh
case "$1" in
pre)
case "$SYSTEMD_SLEEP_ACTION" in
suspend|hibernate)
/usr/bin/nvidia-sleep.sh "$SYSTEMD_SLEEP_ACTION"
;;
suspend-after-failed-hibernate)
/usr/bin/nvidia-sleep.sh "suspend"
;;
esac
;;
post)
/usr/bin/nvidia-sleep.sh "resume"
;;
esac
Following systemd’s documentation, I’ve found out that all the executables in /lib/systemd/system-sleep/ are run before entering sleep mode, and the first argument passed is either “pre” or “post”: therefore, when it’s “post” I simply call nvidia-sleep.sh with the “resume” argument while when it’s “pre” I call it with the value of the SYSTEMD_SLEEP_ACTION variable.
It feels like an hack but it works, should I consider this solved?
2 Likes
Its the only thing I found that works. Its a hack in a sense that it requires re-editing when updating Nvidia drivers, I guess. Thanks for the solution. Long live the hack :D (Seriously, I hope it won’t break with upcoming drivers).
@amrits Sorry to bug you, this issue is still present with driver 515.43.04.
Sounds like this would be a simple fix, if this “hack” is considered good enough.
2 Likes
I had to make one tweak to your code, and now suspend-then-hibernate is 100% working: I had to put sleep 4 after /usr/bin/nvidia-sleep.sh "resume", otherwise my laptop would hang after resuming from hibernation.
I suspect this is because the echo "resume" > /proc/driver/nvidia/suspend and echo "hibernate" > /proc/driver/nvidia/suspend calls (via /usr/bin/nvidia-sleep.sh) occur too rapidly for the nvidia driver to handle properly.
To debug, I put this at the top of /lib/systemd/system-sleep/nvidia:
echo "$(date) $1 $2 $SYSTEMD_SLEEP_ACTION" >> /var/log/my-sleep-log
Before adding the sleep, I saw:
Sun Jun 12 08:58:41 PM CDT 2022 pre suspend-then-hibernate suspend
Sun Jun 12 08:58:53 PM CDT 2022 post suspend-then-hibernate suspend
Sun Jun 12 08:58:54 PM CDT 2022 pre suspend-then-hibernate hibernate
After the sleep I see:
Sun Jun 12 09:28:19 PM CDT 2022 pre suspend-then-hibernate suspend
Sun Jun 12 09:28:31 PM CDT 2022 post suspend-then-hibernate suspend
Sun Jun 12 09:28:36 PM CDT 2022 pre suspend-then-hibernate hibernate
Sun Jun 12 09:29:18 PM CDT 2022 post suspend-then-hibernate hibernate
So it looks like before my tweak the nvidia driver wouldn’t properly prepare for hibernation, causing the system to crash/hang immediately after the first frame of video painted to screen, showing a frozen desktop.
While it would be nice if nvidia would write proper drivers, I would be happy if they at least rolled out this change to /lib/systemd/system-sleep/nvidia so suspend-then-hibernate isn’t broken by default (and so I don’t have to keep manually altering this file after package updates). If support can forward this to the dev team, it would be much appreciated.
1 Like