You are not logged in.

Pages: 1

#1 2025-05-26 14:42:22

raldone01
Member
Registered: 2021-11-12
Posts: 11

[Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

I have an AMD cpu with an integrated 780M.
I frequently connect an NVIDIA egpu.

`vainfo` works fine without the egpu but when the egpu is connected `vainfo` crashes and I get no video accel.
The error message is quite unhelpful.

I have tries all relevant combinations of `LIBVA_DRIVER_NAME` and `VDPAU_DRIVER`.

Both vars unset NO egpu:

❯ vainfo
Trying display: wayland
vainfo: VA-API version: 1.22 (libva 2.22.0)
vainfo: Driver version: Mesa Gallium driver 25.1.1-arch1.1 for AMD Radeon 780M (radeonsi, phoenix, LLVM 19.1.7, DRM 3.61, 6.14.7-arch2-1)
vainfo: Supported profile and entrypoints
      VAProfileH264ConstrainedBaseline:	VAEntrypointVLD
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSlice
      VAProfileH264Main               :	VAEntrypointVLD
      VAProfileH264Main               :	VAEntrypointEncSlice
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileH264High               :	VAEntrypointEncSlice
      VAProfileHEVCMain               :	VAEntrypointVLD
      VAProfileHEVCMain               :	VAEntrypointEncSlice
      VAProfileHEVCMain10             :	VAEntrypointVLD
      VAProfileHEVCMain10             :	VAEntrypointEncSlice
      VAProfileJPEGBaseline           :	VAEntrypointVLD
      VAProfileVP9Profile0            :	VAEntrypointVLD
      VAProfileVP9Profile2            :	VAEntrypointVLD
      VAProfileAV1Profile0            :	VAEntrypointVLD
      VAProfileAV1Profile0            :	VAEntrypointEncSlice
      VAProfileNone                   :	VAEntrypointVideoProc

Both unset WITH egpu:

❯ vainfo
Trying display: wayland
fish: Job 1, 'vainfo' terminated by signal SIGFPE (Floating point exception)

Both set to radeonsi  NO egpu:

LIBVA_DRIVER_NAME=radeonsi
VDPAU_DRIVER=radeonsi

❯ vainfo
Trying display: wayland
vainfo: VA-API version: 1.22 (libva 2.22.0)
vainfo: Driver version: Mesa Gallium driver 25.1.1-arch1.1 for AMD Radeon 780M (radeonsi, phoenix, LLVM 19.1.7, DRM 3.61, 6.14.7-arch2-1)
vainfo: Supported profile and entrypoints
      VAProfileH264ConstrainedBaseline:	VAEntrypointVLD
      VAProfileH264ConstrainedBaseline:	VAEntrypointEncSlice
      VAProfileH264Main               :	VAEntrypointVLD
      VAProfileH264Main               :	VAEntrypointEncSlice
      VAProfileH264High               :	VAEntrypointVLD
      VAProfileH264High               :	VAEntrypointEncSlice
      VAProfileHEVCMain               :	VAEntrypointVLD
      VAProfileHEVCMain               :	VAEntrypointEncSlice
      VAProfileHEVCMain10             :	VAEntrypointVLD
      VAProfileHEVCMain10             :	VAEntrypointEncSlice
      VAProfileJPEGBaseline           :	VAEntrypointVLD
      VAProfileVP9Profile0            :	VAEntrypointVLD
      VAProfileVP9Profile2            :	VAEntrypointVLD
      VAProfileAV1Profile0            :	VAEntrypointVLD
      VAProfileAV1Profile0            :	VAEntrypointEncSlice
      VAProfileNone                   :	VAEntrypointVideoProc

Both set to radeonsi WITH egpu:

LIBVA_DRIVER_NAME=radeonsi
VDPAU_DRIVER=radeonsi

❯ vainfo
Trying display: wayland
libva error: /usr/lib/dri/radeonsi_drv_video.so init failed
vaInitialize failed with error code 2 (resource allocation failed),exit

Both set to nvidia with/without egpu:

LIBVA_DRIVER_NAME=nvidia
VDPAU_DRIVER=nvidia

❯ vainfo
Trying display: wayland
fish: Job 1, 'vainfo' terminated by signal SIGFPE (Floating point exception)

I set the vars via `/etc/environment` and rebooted in between. All tests with the egpu connected have been done with it being the primary wayland device (if it is just used as a copy display bridge it vainfo works fine but performance is really bad).

I am not sure where to report this issue. I think it is an nvidia implementation issue. Do you know a workaround to get vainfo to work with an egpu on wayland?

Last edited by raldone01 (2025-06-01 11:39:23)

Offline

#2 2025-05-27 12:37:56

Lone_Wolf
Administrator
From: Netherlands, Europe
Registered: 2005-10-04
Posts: 13,892

Re: [Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

for clarity :

You do have libva-nivida-driver installed ?

Is the eGPU RTX 2000 or later and if so are you using nivida-open ?


P.S.
Please edit your post to use code tags for such output, see https://bbs.archlinux.org/help.php#bbcode

Last edited by Lone_Wolf (2025-05-27 12:38:47)

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

Online

#3 2025-05-27 13:29:38

raldone01
Member
Registered: 2021-11-12
Posts: 11

Re: [Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

Good points.

:: libva-nvidia-driver-0.0.13-1 and libva-vdpau-driver-0.7.4-7 are in conflict. Remove libva-vdpau-driver? [y/N]

So I had `libva-vdpau-driver` installed.
I will try it with the va driver instead on Thursday. I thought vdpau was better because it supports both va and vdpau.

It is an RTX 4080.
I am using nvidia-open.

local/hip-runtime-nvidia 6.4.0-1
local/lib32-nvidia-utils 570.153.02-1
local/libva-nvidia-driver 0.0.13-1
local/nvidia-open-dkms 570.153.02-1
local/nvidia-utils 570.153.02-1
local/opencl-nvidia 570.153.02-1

Damn bbcode is annoying. Too bad that there is no markdown mode.

Last edited by raldone01 (2025-05-27 13:31:39)

Offline

#4 2025-05-28 11:10:23

Lone_Wolf
Administrator
From: Netherlands, Europe
Registered: 2005-10-04
Posts: 13,892

Re: [Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

I thought vdpau was better because it supports both va and vdpau.

On X11 maybe.

for wayland : look were vdpau is listed in https://forums.developer.nvidia.com/t/w … ies/333827

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

Online

#5 2025-05-29 13:42:10

raldone01
Member
Registered: 2021-11-12
Posts: 11

Re: [Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

Thanks for the link.
Good to know.

You fixed my issue!
Thank you so much!

Installing the `libva-nivida-driver` package fixed my hardware video acceleration.


I have another follow up question.
If you don't know I will just make a new thread.

So when I boot up my laptop with the egpu plugged in it isn't detected. I have to replug it every time late in the boot process.

I tried to early load the relevant drivers but it didn't help:

❯ cat /etc/mkinitcpio.conf
...
MODULES=(tpm_crb hid usbhid usbcore thunderbolt usbnet typec crypto_user crypto_simd cryptd btrfs nvidia_drm nvidia nvidia_uvm amdgpu radeon)
...
HOOKS=(base systemd autodetect microcode modconf kms keyboard sd-vconsole block sd-encrypt lvm2 filesystems fsck)
...

Offline

#6 2025-05-29 15:25:12

Lone_Wolf
Administrator
From: Netherlands, Europe
Registered: 2005-10-04
Posts: 13,892

Re: [Solved] [EGPU] `'vainfo' terminated by signal SIGFPE`

So when I boot up my laptop with the egpu plugged in it isn't detected. I have to replug it every time late in the boot process.

That requires other troubleshooting steps and should be dealt with in a new thread in Kernel & Hardware board.

In the new thread link to this thread and add lspci -k output with and without the egpu plugged in .
Also add a full journal output from boot wuth the egpu connected, then removing / replugging it.
As root or with root rights run 

# journalctl -b | curl -F 'file=@-' 0x0.st

to upload the journal to  a public hosting site.
That cmmand will output a link, post that link.

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

Online

Pages: 1

Board footer

Powered by FluxBB