You are not logged in.
Hi, I noticed that when I upgrade to kernel 4.15.4-zen from testing, instead of seeing nvidia-modeset and irq/27-nvidia in my htop, I get a single Kworker which uses about a full CPU. I also noticed that the system uses Xwayland instead of Xorg -- I am using wayland either way, but I guess it starts the session via Xorg. I am using the nvidia-driver 390.25-11. Even though it is working now with the 4.15.3 kernel, I see messages like this:
[ 22.285] (WW) NVIDIA(0): Suspend swap group failed.
[ 22.285] (WW) NVIDIA(0): Resume swap group failed.
On the 4.15.4 kernel, I would get flickering, and the system would become unresponsive, then I'd get a message about kernel error in drm_kms_helper, flip_done timed out in dmesg.
Last edited by afader (2018-03-21 01:15:20)
Offline
Offline
Thank you, but the strange thing here is that it works great with 4.15.3 and broken on 4.15.4. So regression?
Offline
You could bisect between 4.15.3 and 4.15.4 and see if you can identify the bad commit or if just rebuilding the same sources affects the result.
Offline
4.15.4
...
4.15.4-zen
...
4.15.3
You should be very precise on which kernels are actually used. All zen? Did you try the vanilla ones?
Online
All zen?
Yes, I'm sorry, I am specifically using the 4.15.3-zen now. I didn't test to see if the 4-ARCH or 3-ARCH had the same discrepancy.
Offline
It would certainly help to test the plain 4.15.4 behavior.
Online
I almost thought I tackled the issue. As long as I do not switch to a virtual console,
the suspend/resume messages won't show (for now at least).
The nvidiafb module does not seem to be causing the issue, loaded or not. Switching to any virtual console triggers it for me.
heftig changed the kernel config to be more Ubuntu like, picked some of their kernel config options
and introduced nvidiafb as a side effect.
Here's heftig's commit (package linux-4.15.3-1 is without nvidiafb and >= linux-4.15.3-2 includes nvidiafb as module)
https://git.archlinux.org/svntogit/pack … 069c84aaec
To be precise he included blacklist options for nvidiafb / rivafb / nouveau in /usr/lib/modprobe.d/nvidia.conf
with package >= nvidia-390.25-11 for kernel 4.15.3-2.
Normally in the past, having both the nvidia binary and the nvidiafb modules caused a few messages
about the unsupported status and incompatibility issues in the Xorg log files.
That is why the new blacklist options were used in combination with the nvidia binary blob.
Also the journal shows another message when the nvidiafb module gets loaded, but seems to be a different issue:
kernel: nvidiafb: cannot request PCI regions
Last edited by NiceGuy (2018-02-19 18:17:34)
Offline
Good find @NiceGuy. It definitely has something to do with nvidiafb. Also, it only affects Wayland, Xorg seems OK.
Offline
Edit: Nevermind. Yes, that is the problem, those ubuntu config vars
Last edited by afader (2018-03-04 04:01:12)
Offline