Arch Linux

Jastix

Member

Registered: 2022-01-23

Posts: 8

[SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

I've been using Arch more than half a year already, and everything was ok, but recently I got some new hardware, though not the video card. So I reinstalled the system and did all the necessary steps same as in previous installation. I also did stuff that https://wiki.archlinux.org/title/NVIDIA advises, but at the moment I launch the game, I get these errors in the journal

Oct 31 18:33:34 XorS kernel: [drm:nv_drm_fence_context_create_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000700] Failed to allocate fence signaling event
Oct 31 18:33:35 XorS kernel: [drm:nv_drm_fence_context_create_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000700] Failed to allocate fence signaling event
Oct 31 18:34:09 XorS assert_20221031183409_32.dmp[2585]: Uploading dump (out-of-process)
                                                         /tmp/dumps/assert_20221031183409_32.dmp
Oct 31 18:34:09 XorS audit[1361]: ANOM_ABEND auid=1000 uid=1000 gid=984 ses=1 pid=1361 comm="steam" exe="/home/jastix/.local/share/Steam/ubuntu12_32/steam" sig=11 res=1
Oct 31 18:34:09 XorS crash_20221031183409_33.dmp[2589]: Uploading dump (out-of-process)
                                                        /tmp/dumps/crash_20221031183409_33.dmp

I was searching for the information about my issue, but seems like there is not much about it. The most common advice I could find is to set "PRIME Synchronization" to 0 for those who has hybrid graphics. But it's not my case.
I thought it could also be some bug in the driver, but within last few days there were two driver updates, but the problem stays.

Eventually I figured out that removal of nvidia_drm.modeset=1 from kernel parameters stops the issue. At least the error doesn't show up at the game start, but I didn't test much longer, in the case it will still happen while the game is running.

So I have two questions:
1. What can I do to try to solve this issue when nvidia drm modeset is ON? I deliberately mentioned that the video card is the same, because on the previous installation I had no issues with this parameter ON.
2. What issues can I face with drm modeset OFF? Will it be lower fps in games or inability to use some soft for video processing? I also tried to find the information about it, but couldn't find something specific. Only general words like "without this you will get only some basic functionality" or "just better have it ON".

Last edited by Jastix (2022-11-06 00:27:57)

sukulent

Member

Registered: 2022-09-20

Posts: 5

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

Hey, after the launch of kernel 6 I have started getting this error too.
I am using Lenovo Legion laptop with integrated intel and dedicated nvidia - my system crashes after I offload 3D application to the nvidia gpu.

I have tried whole bunch of nvidia driver and kernel versions, with dkms and without. Currently running nvidia/520.56.06 and 6.0.6-zen1-1-zen, I have set-up nvidia prime render offload as in the nvidia docs and arch wiki, using  intel modesetting driver and having nvidia_drm.modeset=1 as that's required for prime offload to work.

Here is some information about the nvidia_drm.modeset=1 setting https://forums.developer.nvidia.com/t/u … g/204068/2

I could try turning off the integrated intel gpu in bios and testing if nvidia_drm.modeset=0 doesn't crash
EDIT: tried running only on the nvidia gpu and it doesn't crash. Confirmed.

I will be happy to provide more information, but I don't want to bloat this reply.

Last edited by sukulent (2022-11-02 11:52:44)

d.ALT

Member

Registered: 2019-05-10

Posts: 920

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

@Jastix, @sukulent:
Did you try LTS? Can you both do some tests?:

pacman -Syu nvidia-lts linux-lts

---

<49,17,III,I>    Fama di loro il mondo esser non lassa;
<50,17,III,I>    misericordia e giustizia li sdegna:
<51,17,III,I>    non ragioniam di lor, ma guarda e passa.

sukulent

Member

Registered: 2022-09-20

Posts: 5

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

@Jastix, @sukulent:
Did you try LTS? Can you both do some tests?:

pacman -Syu nvidia-lts linux-lts

Yes, I tried linux-lts with nvidia-lts and with nvidia-dkms, I also tried multiple older versions of nvidia-dkms using the downgrade utility, with different kernels - I didn't find any combination which wouldn't crash after a few minutes in Metro Lasta Light redux (native on steam).

I also tried downgrading mesa, but that did nothing, as expected. The only way I got rid of crashes was to disable intel gpu in bios and run only on nvidia.

I have gone through journalctl now, looking for more error messages:

around the time of the crash there is at least one

kernel: [drm:nv_drm_fence_context_create_ioctl [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to allocate fence signaling event

also I found (only) one

kernel: [drm:drm_new_set_master] *ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to grab modeset ownership

also found a few

kernel: nvidia-gpu 0000:01:00.3: i2c timeout error e0000000
kernel: ucsi_ccg 11-0008: i2c_transfer failed -110
kernel: ucsi_ccg 11-0008: ucsi_ccg_init failed - -110

but idk if that's related, shows up without a crash.

Jastix

Member

Registered: 2022-01-23

Posts: 8

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

@Jastix, @sukulent:
Did you try LTS? Can you both do some tests?:

pacman -Syu nvidia-lts linux-lts

I tried it out now, sadly the error stays.
I also sometimes get the same error like sukulent has:

Failed to grab modeset ownership

But this issue shows up pretty randomly, so I couldn't find any pattern to reproduce it at will.

Tbh I'm not sure at what moment the error begun to show up, was it since I reinstalled the system or after the installation of some programs. I sometimes was checking the journal for errors and fails but only after booting the system, not after some usage. My bad. Nvidia drm error doesn't crash the game neither it makes it unplayable, it just drops performance significantly and only in some areas in the game. I naturally got used to performance drops with my old hardware, so I could not pay attention at the beginning. Only after a while I realized, that the new hardware just gives me worse performance in the situations where even my old CPU was easily providing 120 FPS (I play Planetside 2 which is very CPU bound).

I think if I won't find the solution until this weekend, I'll just reinstall the system, put only basic stuff like DE, nvidia driver and steam and try it out on the clean system.

Jastix

Member

Registered: 2022-01-23

Posts: 8

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

The clean system re-installation didn't help but the issue is actually solved the other way. I removed nvidia-520 driver and installed nvidia-515 from AUR. No errors neither performance drops so far with modeset=1.

sukulent

Member

Registered: 2022-09-20

Posts: 5

Re: [SOLVED]nvidia_drm.modeset=1, Failed to allocate fence signaling event

The clean system re-installation didn't help but the issue is actually solved the other way. I removed nvidia-520 driver and installed nvidia-515 from AUR. No errors neither performance drops so far with modeset=1.

So I now tried the "nvidia-515" from AUR - uninstalled useful things like cuda, had to ignore dependencies as the only 515 driver from AUR is called "nvidia-dkms-6" which goes with "nvidia-utils-6", but depends on "nvidia-utils"

I have played minecraft with lot of forge mods and it crashed once after something like two hours. I did not find any of the errors I have found before on the other versions of nvidia driver.

One strange thing happening sometimes tho: The screen freezes, sound is fine, everything is running fine probably, to "fix" I have to switch to a console and back, "refreshing" the GPU to start displaying again. Hybrid graphic shenanigans? And nvidia driver still crashing, now without errors in logs.

Probably should start new topic on the issue when this is marked as SOLVED.