You are not logged in.
Hello, I am aware this is a known problem but all the solutions I have tried so far have not worked. First Let me describe the symptoms:
Arch boots, really fast then SDDM starts and it hangs there. I cannot access other TTYs or anything, I can only reboot. Once I reboot usually the process repeats and then SDDM starts correctly. This is a hit-and-miss situation because sometimes it boots normally and sometimes it doesn't.
Here is what I've tried:
- I have read somewhere that it might be because of the boot speed is too fast (it is!) so SDDM tries to start before the NVIDIA driver is fully loaded: To remedy this I attempted to add a systemd unit file that delays the startup as indicated here . This has actually had mixed results because it works most of the time, but sometimes I had the system freeze anyways with the following error BEFORE SDDM starts:
nmi watchdog bug soft lockup - cpu#0 stuck for 22sI have read that this might be caused by a faulty PSU, but honestly I doubt it because: I don't have this problem with any other distro (fedora, Debian or Manjaro were installed recently and they booted fine with the propietary driver, the only distros that do this are Antergos and Arch). Moreover when SDDM starts there is no error even after prolonged periods of time (a day of uptime for example)
- I have considered that maybe the nouveau driver is popping up and loading . So I blacklisted it in /etc/modprobe.d ( added a .conf file with the contents "blacklist nouveau"), I also added nouveau.modeset = 0 to kernel boot parameter.
- I have thought that perhaps nvidia propietary driver modesetting should be configured so I added (nvidia nvidia_modeset nvidia_uvm nvidia_drm) to the MODULES section on /etc/mkinitcpio.conf (and regenerated the initramfs). I also added the nvidia.modeset = 1 parameter to grub.
I can't really see any relevant info on the logs regarding startup (and since I cannot switch TTYs I can't see journalctl -xe live when the crash happens). Any idea on how to get better logs would be appreciated.
After all these attempts this still fails and I am running out of ideas. One of the things I think I could try is to see what exactly Manjaro does with the configuration and try to mimic that, but that would mean that I would have to install Manjaro and reinstall Arch and I am kinda happy with my current install (this being the only problem).
I also considered rolling without SDDM but my girlfriend ocassionally uses the computer.
So I apologize in advance if this is solved somewhere else, but I really think I tried everything I could find in these forums and in the wiki.
Thanks.
Edit: my video card is a NVIDIA 1050 GTX.
Last edited by juanfgs (2018-11-02 23:26:36)
Offline
If you've added your modules to your mkinitcpio most worries wrt to SDDM loading too fast should be alleviated even without adding an artificial delay. FWIW nvidia.modeset = 1 would be incorrect, it should be nvidia-drm.modeset=1. Simply from similar symptoms I'm leaning towards the starved random seed, however then pressing buttons should eventually resolve the situation (however there is a bug with the prop drivers and not having modeset properly enabled that will prevent you from loading a TTY so maybe that's what you're seeing in that case). Try and install and enable haveged.
If that doesn't fix it,can you get us a full log, after restarting after a crash post
sudo journalctl -b-1that way you can get the entire log from the previous (crashing) boot. (Also future disclaimer: When posting logs to the boards or bug reports, don't add the -x option, it adds unnecessary fluff for people that are familiar with reading the outputs and might drown actual things of relevance)
Offline
Thanks for the response, I am currently at work but I will try that and report back as soon as I get back home.
Simply from similar symptoms I'm leaning towards the starved random seed, however then pressing buttons should eventually resolve the situation (however there is a bug with the prop drivers and not having modeset properly enabled that will prevent you from loading a TTY so maybe that's what you're seeing in that case)
not sure if I am following what you are saying. you mean pressing buttons when the screen is blank?
don't add the -x option, it adds unnecessary fluff for people that are familiar with reading the outputs and might drown actual things of relevance
TIL, thanks a lot!
can't wait to get home and try all this.
Offline
yes, anything else that isn't switching to a TTY (because as mention that happens to break as well, but for different reasons), see this this earlier thread for when the problem first surfaced for some more information.
Last edited by V1del (2018-11-02 15:16:39)
Offline
Hello I think I solved it (kinda).
Turns out that SDDM and NVIDIA had nothing to do with it! (I actually had set nvidia-drm.modeset = 1 properly just wrote it wrong here! )
I turned off sddm and tried to reproduce the problem, and suddenly the console started to spit :
nov 02 19:10:58 leela kernel: INFO: rcu_sched detected stalls on CPUs/tasks:
nov 02 19:10:58 leela kernel: 2-....: (5 GPs behind) idle=6e6/0/1 softirq=526/526 fqs=24002 last_accelerate: be40/d789, nonlazy_posted: 0, L.
nov 02 19:10:58 leela kernel: (detected by 1, t=72007 jiffies, g=-76, c=-77, q=2)
nov 02 19:10:58 leela kernel: Sending NMI from CPU 1 to CPUs 2:
(...)
watchdog: BUG: soft lockup - CPU#2 stuck for 23s! [swapper/2:0]googled it and I found out that a lot of people are having that trouble with the 4.18.x branch of the kernel. (that's why it works on manjaro since it has the 4.17 branch!)
So I installed linux-lts linux-lts-headers nvidia-lts and rebooted.
The system starts up perfectly now, and it is incredibly fast (systemd-analyze gives 6 seconds to userspace). Now this is not a proper solution if someone wants to use the 4.18 branch, but honestly I am not a kernel/module dev so it is not a big deal for me.
Thanks V1del for the help, it guided me to the right way ![]()
Dunno how can I add [Solved] to the thread title
Last edited by juanfgs (2018-11-02 22:40:05)
Offline
If this is the issue you are refering to it was fixed in 4.18.9.
You can mark your thread as solved by editing the first post and prepending [SOLVED] to the title.
Offline
If this is the issue you are refering to it was fixed in 4.18.9.
You can mark your thread as solved by editing the first post and prepending [SOLVED] to the title.
Would it be useful to somebody to report this still happening in 4.18.16 with my current CPU?
Offline
Arch is probably going to switch to 4.19 rather than do a 4.18.17 release. So I would suggest checking if the issue is still present in 4.19.arch1-1 which is in testing. You would also need the matching nvidia package from testing.
Last edited by loqs (2018-11-02 23:32:49)
Offline
I will try whenever I get some time(kinda busy weekend), for now I think I will live with LTS.
Last edited by juanfgs (2018-11-02 23:37:07)
Offline