Plex LXC on a Intel NUC (Geekom Mini IT13) : /dev/dri/card0 change to /dev/dri/card1 after reboot

[MilesTEG1]

Hello,
I'm using the LXC Plex script installation, it works great, but I'm confronted to a problem with my NUC, a Geekom Mini IGT13, with Intel Core i9-13900H (Raptor Lake), and iGPU Intel Iris Xe 1.50 GHz.
The /dev/dri/card0 change to /dev/dri/card1 after a reboot of the proxmox host, and change back to /dev/dri/card0 after another reboot, and etc...
It's annoying, because Plex LXC won't start if the device is /dev/dri/card1 and set to /dev/dri/card0 in the LXC ressources section.

Do you have a workaround for me?
I tried to create a link inside de LXC but it's not persistent...

Here what I have on the proxmox host:

After a reboot, only the card0 change to card1, the by-path/pci-0000:00:02.0-card keep its name, but point to card1.

Thanks in advance.

[XmemphistoX]

I dont have the same iGPU as you, I'm on alderlake but in my LXC i only link the renderer and not the video. HW transcoding works for me doing it that way.

[MilesTEG1]

Hello,
I don't know if /dev/dri/card0 is needed or not.
The LXC Script for creation of Plex LXC set the /dev/dri/card0 in the Ressources section.
@tteck What do you think?
Have you a workaround to fix the /dev/dri/card0 in the proxmox host? or to get a link inside the LXC in en event of a change from /dev/dri/card0 to /dev/dri/card1 ?

Thanks in advance.

edit : I just try to remove /dev/dri/card0 from the ressources allowed to the LXC. And Plex keep transcode with HW acceleration.
So it may no be needed... Or I don't understand very much how it's working...

[Mellowlynx]

@MilesTEG1 On a side note:
What shell are you using, looks like an Oh My Zsh Theme, but way better :)
Would love to know what you are using.

[MilesTEG1]

Hi @Mellowlynx
I use ZSH + OhMyZSH + Powerlevel10K.
It took me some times to achieve that look for my prompt ;)
Have a look here : https://github.com/MilesTEG1/zsh_my_prompt_customization
I just updated the macOS .zshrc and add one for ProxmoxVE / Linux.
Feel free to comment or add something you think will add value.

[Mellowlynx]

Thanks, I will take a look :)

[tteck]

@MilesTEG1
Within the Proxmox Shell, execute nano /etc/pve/lxc/<CTID>.conf change <CTID> to the Plex container ID, then change card0 to card1 or vice versa. Reboot the LXC

[MilesTEG1]

@tteck Hello
Thank for your answer, but it doesn't solve my problem.
I already did that, through proxmox WebUI. I know how to change card0 to card1 or vice-versa.

My demand was about to not have to do this manually.

• by fixing the card0 on the proxmox host, so the LXC can use always card0.
• by setting the lxc to create a link for card1 -> card0 if card0 doesn't exist and card1 exist.

Do you know if one or the other is possible? And how?

[oliv99]

ls -al /dev/dri
drwxr-xr-x 3 root root 120 Jun 19 20:07 .
drwxr-xr-x 20 root root 5180 Jun 19 20:06 ..
drwxr-xr-x 2 root root 80 Jun 19 20:06 by-path

either
crw-rw---- 1 root video 226, 0 Jun 19 20:06 card0
or
crw-rw---- 1 root video 226, 1 Jun 19 12:38 card1

crw-rw---- 1 root render 226, 128 Jun 19 20:06 renderD128

pct config xxx

lxc.cgroup2.devices.allow: a
lxc.cap.drop:
lxc.cgroup2.devices.allow: c 226:0 rwm
lxc.cgroup2.devices.allow: c 226:1 rwm
lxc.cgroup2.devices.allow: c 226:128 rwm
lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir
lxc.mount.entry: /dev/dri/card0 dev/dri/card0 none bind,optional,create=file
lxc.mount.entry: /dev/dri/card1 dev/dri/card1 none bind,optional,create=file
lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file

I have the same issue on N6005 (may be not an issue..), then i have added both card0 and card1 in the LXC config file
no problem to start the LXC

[MilesTEG1]

pct config xxx

lxc.cgroup2.devices.allow: a lxc.cap.drop: lxc.cgroup2.devices.allow: c 226:0 rwm lxc.cgroup2.devices.allow: c 226:1 rwm lxc.cgroup2.devices.allow: c 226:128 rwm lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir lxc.mount.entry: /dev/dri/card0 dev/dri/card0 none bind,optional,create=file lxc.mount.entry: /dev/dri/card1 dev/dri/card1 none bind,optional,create=file lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file

How do you manage to have this output to the pct config xxx command ?

Mine is:

arch: amd64
cores: 6
description: <div align='center'><a href='https%3A//Helper-Scripts.com' target='_blank' rel='noopener noreferrer'><img src='https%3A//raw.githubusercontent.com/tteck/Proxmox/main/misc/images/logo-81x112.png'/></a>%0A%0A  # Plex LXC%0A%0A  <a href='https%3A//ko-fi.com/D1D7EP4GF'><img src='https%3A//img.shields.io/badge/&#x2615;-Buy me a coffee-blue' /></a>%0A  </div>%0A
dev0: /dev/dri/card0,gid=44
dev1: /dev/dri/renderD128,gid=104
features: mount=cifs,nesting=1
hostname: Plex-LXC
memory: 4096
net0: name=eth0,bridge=vmbr0,hwaddr=F2:6F:7A:EC:1D:D8,ip=dhcp,type=veth
onboot: 1
ostype: debian
rootfs: local-lvm:vm-250-disk-0,size=52G
startup: order=4
swap: 1024
tags: PMS;LXC;Debian;proxmox-helper-scripts

I have the same issue on N6005 (may be not an issue..), then i have added both card0 and card1 in the LXC config file
no problem to start the LXC

How do you manage to get the LXC to start when card0 and card1 are present in the Ressources tab, and when one of the two is missing?

[MilesTEG1]

I try to add /dev/dri/card1 through WebUI, but I get this error:

I edited with nano in CLI, and stop de LXC. Then I tried to start it:

[oliv99]

I am using a privileged container
I do not use the proxmox GUI to add device
I edit manually the config file and add the relevant lines

ls -l /etc/pve/lxc/
nano /etc/pve/lxc/***.conf (replace *** with your CTID)


[MilesTEG1]

@oliv99 My LXC is privileged too.
I try to add the card1 :

Then start the LXC :

As you can see, it doesn't work if /dev/dri/card1 doesn't exist.

[MilesTEG1]

try

nano /etc/pve/lxc/250.conf (250 is your CTID)

Replace

dev0: /dev/dri/card0,gid=44 dev1: /dev/dri/renderD128,gid=104

by

lxc.cgroup2.devices.allow: a lxc.cap.drop: lxc.cgroup2.devices.allow: c 226:0 rwm lxc.cgroup2.devices.allow: c 226:1 rwm lxc.cgroup2.devices.allow: c 226:128 rwm lxc.mount.entry: /dev/dri dev/dri none bind,optional,create=dir lxc.mount.entry: /dev/dri/card0 dev/dri/card0 none bind,optional,create=file lxc.mount.entry: /dev/dri/card1 dev/dri/card1 none bind,optional,create=file lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file

no guarantee

Oh ! This seems to be working.
I'll have to try a reboot of the proxmox host to see ^^

[MilesTEG1]

@oliv99 I rebooted the proxmox host. As expected, /dev/dri/card0 changed to /dev/dri/card1.
But your modifications in the old ways, with lxc.mount... let my LXC start !

But, as @XmemphistoX said, it seems the Plex doesn't need the /card0 or card1 to transcode.

That's a question I'll ask to @ChuckPa in the plex forums.

[chenks]

@oliv99 I rebooted the proxmox host. As expected, /dev/dri/card0 changed to /dev/dri/card1. But your modifications in the old ways, with lxc.mount... let my LXC start !

But, as @XmemphistoX said, it seems the Plex doesn't need the /card0 or card1 to transcode.

That's a question I'll ask to @ChuckPa in the plex forums.

did you find out if it was needed?
i encountered the same thing yesterday.

reboot PVE and card0 had changed to card1 meaning both Scrypted and Plex LXCs failed to start.
if i manually changed the config to card1 then both started fine.
however i then rebooted PVE again, and it had changed back to card0, meaning Scrypted and Plex both failed to start again until i manually changed it back to card0.

ps, i don't have Plex Pass so the hardware transcoding option within plex isn't available to me anyway (and the devices i use can direct play all content).

[MilesTEG1]

@oliv99 I rebooted the proxmox host. As expected, /dev/dri/card0 changed to /dev/dri/card1. But your modifications in the old ways, with lxc.mount... let my LXC start !
But, as @XmemphistoX said, it seems the Plex doesn't need the /card0 or card1 to transcode.
That's a question I'll ask to @ChuckPa in the plex forums.

did you find out if it was needed? i encountered the same thing yesterday.

reboot PVE and card0 had changed to card1 meaning both Scrypted and Plex LXCs failed to start. if i manually changed the config to card1 then both started fine. however i then rebooted PVE again, and it had changed back to card0, meaning Scrypted and Plex both failed to start again until i manually changed it back to card0.

ps, i don't have Plex Pass so the hardware transcoding option within plex isn't available to me anyway (and the devices i use can direct play all content).

According to PlexPa in Plex Forums, /dev/dri/card0 is needeed for ToneMapping.

This name chaging of card0 to card1 and vice-versa, is verry anoying...
I didn't find a solution. It may be a kernel problem, but I don't understand very much of this.

[chenks]

According to PlexPa in Plex Forums, /dev/dri/card0 is needeed for ToneMapping.

which is a Plex Pass feature anyway? so if you don't have Plex Pass then it's not needed?

This name chaging of card0 to card1 and vice-versa, is verry anoying...
I didn't find a solution. It may be a kernel problem, but I don't understand very much of thi

the switching seems to be a proxmox issue rather than a script problem, has anyone asked the question in proxmox forums/reddit ?

[ChuckPa]

Plex's PA office here ;)

1. renderD128 / renderD129 / ... is for the transcoding (encoding & bitrate conversion) part of the process.
2. card0 / card1 / ... is how we access the openCL layer on Intel CPUs with Quick Sync Video to perform tone mapping of HDR.
3. With Nvidia, we use the CUDA layer directly through their API.

In Setting - Server - Transcoding - Show Advanced, you can change the trancoding device (engine) from "Auto" to the specific device you want to use. This is how you can select an Nvidia (for example) on a machine with both Quick Sync Video and Nvidia card

[ChuckPa]

Supplemental:

I just remembered but don't know where it's documented. (Somewhere in the Plex forums)
I had to do it when setting up my Proxmox test box

You have to blacklist one of the video drivers to get the passthrough right (card0 + renderD128)

For me, on my Ubuntu host, LXD/LXC is trivial.

1. Host gets the drivers (if needed)
2. lxc config to pass into the container:

[chuck@lizum lxc.2003]$ cat add-gpu

#
# Argument #1 is LXC container to add

[ -z "$1" ] && "Error: missing container name" && exit 1

if [ "$(lxc list | grep "$1")" == "" ]; then
  echo "Error: Unknown container name '$1'"
  exit 2
fi

# Make certain /dev/dri/renderD128 exists
if [ ! -e /dev/dri/renderD128 ]; then
  echo Error:  This host does not have hardware transcoding ability /dev/dri/renderD128
  exit 3
fi

Gid="$(stat -c %g /dev/dri/renderD128)"

# Add it


 lxc config device add "$1" gpu gpu gid=$Gid
[chuck@lizum lxc.2004]$ 

[MilesTEG1]

Hello @ChuckPa

You have to blacklist one of the video drivers to get the passthrough right (card0 + renderD128)

I use a mini PC with no graphic card except the one included in the CPU Intel Core i9-13900H.
I have no other graphic card to ban at the proxmox startup 😓

Before speaking about the code you posted, may you clarify something for me (and others), please? Does Plex can use either /dev/dri/card0 or /dev/dri/card1 to get the tonemapping done correctly?
You'll see why I ask you that in the next lines.

As for the code you posted, it is a script, right? Which is launched at startup of the proxmox host?
Regardless if it is or not, it gave me an idea to write a script which is launched after proxmox startup (yet to define how to achieve it) to change the container ressource passed to the LXC.
If /dev/dri/card0 it will be passed to the LXC? but if not, but /dev/dri/card1
exists, this one will be passed.
I'll try to wrote the script for it. But I'll do it only for a system with only one graphic card: an embedded one in the CPU, called iGPU.

but before I'll go to my VSCode app to code something for it, @ChuckPa, I need the confirmation that Plex can use either /dev/dri/card0 or /dev/dri/card1 to get the tonemapping done correctly?

Thanks in advance 😃

[ChuckPa]

If you're using Intel QSV (which I did on my NUC7i3BNH), you'll only need add it to the config of the container.

As I showed above,

lxc config device add "$1" gpu gpu gid=$Gid

is the heart of what is done to enumerate it into the container.
Gid comes from the host and you must ensure group membership is correct inside the container.
(Yes, the container runs as root but you still must add the group to avoid getting root-squashed to nobody
-- OR -- run privileged which carries some risks )

[ChuckPa]

Plex looks for card0 and renderD128 (which are how the pair is enumerated by the normal kernel.
card0 is how it accesses OpenCL. renderD128 is the QSV ASIC.

IF you have a stray secondary driver in the host, that's likely how you're getting card0, card1, and renderD128

This is my Nvidia host.
The CPU & mobo have a VGA adapter. That VGA adapter must be blacklisted for things to align properly in my case.

[chuck@glockner ~.1999]$ ls -la /dev/dri
total 0
drwxr-xr-x  3 root root        120 Jun 12 01:30 ./
drwxr-xr-x 20 root root       5940 Jun 12 23:56 ../
drwxr-xr-x  2 root root        100 Jun 12 01:30 by-path/
crw-rw----  1 root render 226,   0 Jun 12 01:30 card0
crw-rw----  1 root render 226,   1 Jun 12 01:30 card1
crw-rw----  1 root render 226, 128 Jun 12 01:30 renderD128
[chuck@glockner ~.2000]$ 

Always remember 'renderD128' = 'card0' + 128 (bit 7 set) -- that's how the enumeration works.

[tteck]

the container runs as root but you still must add the group to avoid getting root-squashed to nobody

The script sets the correct groups for both privileged and unprivileged Hardware Acceleration

Proxmox/install/plex-install.sh

Lines 41 to 45 in 6e6a86a

	if [[ "$CTTYPE" == "0" ]]; then
	sed -i -e 's/^ssl-cert:x:104:plex$/render:x:104:root,plex/' -e 's/^render:x:108:root$/ssl-cert:x:108:plex/' /etc/group
	else
	sed -i -e 's/^ssl-cert:x:104:plex$/render:x:104:plex/' -e 's/^render:x:108:$/ssl-cert:x:108:/' /etc/group
	fi

[tteck]

Before speaking about the code you posted, may you clarify something for me (and others), please? Does Plex can use either /dev/dri/card0 or /dev/dri/card1 to get the tonemapping done correctly?

HDR tone mapping is NOT currently working with the current Proxmox kernel. This must be disabled to get Hardware Acceleration working. #3162

[MilesTEG1]

@ChuckPa: If I understand correctly, you are saying if I pass /dev/dri/card1 to the LXC, it will not work?

Because, that's what I intend to write in the script.

And the latest answer you wrote tells me you seem to not correctly understand my problem...

For a reason I ignore, Proxmox change the /dev/dri/card0 to /dev/dri/card1 after a reboot, and after another reboot, /dev/dri/card1 would be again /dev/dri/card0.
it is not that I have two graphic cards! I only have the iGPU on my Intel Core i9-13900H.

@tteck

HDR tone mapping is NOT currently working with the current Proxmox kernel. This must be disabled to get Hardware Acceleration working.

On the Plex Forum, it was said that this issue was resolved, am I mistaken @ChuckPa?

because when I set this option in Plex preferences, I still get HW acceleration.

[tteck]

For a reason I ignore, Proxmox change the /dev/dri/card0 to /dev/dri/card1 after a reboot, and after another reboot, /dev/dri/card1 would be again /dev/dri/card0.

I have not experienced this behavior.

[tteck]

If you reboot again, is it card0 or card1? It should stay one or the other.

[tteck]

On the Plex Forum, it was said that this issue was resolved, am I mistaken @ChuckPa?

That would be good to know!

[MilesTEG1]

If you reboot again, is it card0 or card1? It should stay one or the other.

Before kernel 6.8.8-2-pve, it switch after a reboot from card0 to card1, or from card1 to card0.

But!!!
Since several minutes, I do a apt update && apt upgrade, and there was a new kernel: 6.8.8-2-pve
And after at minima 6 reboot, /dev/dri/card0 still be card0!
Hourra!

So this was a kernel bug?

[tteck]

To clarify, the Plex Media Server LXC script, whether privileged or unprivileged, assigns the card to either card0 or card1, depending on which is present. It also sets the correct groups. This setup has worked for many users. The card flip-flopping during reboot is uncommon.

[MilesTEG1]

To clarify, the Plex Media Server LXC script, whether privileged or unprivileged, assigns the card to either card0 or card1, depending on which is present. It also sets the correct groups. This setup has worked for many users. The card flip-flopping during reboot is uncommon.

The card flip-flopping (I like very much this expression) was a bug, a kernel bug! Because, since I update to 6.8.8-2-pve, no more flip-flopping. 😄

[tteck]

Now, is HDR tone mapping working?

[tteck]

Previously, if HDR tone mapping was selected, it would stop Hardware Acceleration.

[ChuckPa]

Previously, if HDR tone mapping was selected, it would stop Hardware Acceleration.

We've changed the transcoder.

If it can't tonemap in HW, it will proceed AS IF OFF.

[tteck]

Nice!

[MilesTEG1]

@ChuckPa How can I see if the tonemapping is working fine, in the log?

[ChuckPa]

Go into the PMS log.

tonemap=mobious will be in the FFMPEG invocation command line... AND .. you will not see OpenCL errors in the first first statements beneath it.

Also, from the GUI, you will have TWO transcodes showing for the same item. (the code was developed oddly... we want to fix that)

[MilesTEG1]

Go into the PMS log.

tonemap=mobious will be in the FFMPEG invocation command line... AND .. you will not see OpenCL errors in the first first statements beneath it.

Also, from the GUI, you will have TWO transcodes showing for the same item. (the code was developed oddly... we want to fix that)

Hi @ChuckPa
Before going to bed, I ran a session by forcing transcode on my PMS.
Here a portion of the log:

Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Starting a transcode session afr7629imvh3hph72qeh2j39 at offset 30.0 (state=3)
Jul 01, 2024 23:21:38.932 [128682190449464] ERROR - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Error creating directory "/transcode": Permission denied
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] TPU: hardware transcoding: using hardware decode accelerator vaapi
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] TPU: hardware transcoding: zero-copy support present
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] TPU: hardware transcoding: using zero-copy transcoding
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] [Universal] Using local file path instead of URL: /data/videos/zz-Vidéos-Tests/Sony Swordsmith HDR UHD 4K Demo.mp4
Jul 01, 2024 23:21:38.932 [128682190449464] INFO - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Preparing driver icr for GPU Raptor Lake-P [Iris Xe Graphics]
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39/DriverDL/icr] Skipping download; already exists
Jul 01, 2024 23:21:38.932 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Codecs: hardware transcoding: testing API vaapi for device '/dev/dri/renderD128' (Raptor Lake-P [Iris Xe Graphics])
Jul 01, 2024 23:21:38.935 [128682190449464] INFO - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Preparing driver imd for GPU Raptor Lake-P [Iris Xe Graphics]
Jul 01, 2024 23:21:38.935 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39/DriverDL/imd] Skipping download; already exists
Jul 01, 2024 23:21:38.936 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] TPU: hardware transcoding: final decoder: vaapi, final encoder: vaapi
Jul 01, 2024 23:21:38.936 [128682190449464] ERROR - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Error creating directory "/transcode": Permission denied
Jul 01, 2024 23:21:38.936 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39/JobRunner] Job running: EnableAIL=0 FFMPEG_EXTERNAL_LIBS='/var/lib/plexmediaserver/Library/Application\ Support/Plex\ Media\ Server/Codecs/ad47460-ffe81d9cd51bd27cb3fbbe09-linux-x86_64/' LIBVA_DRIVERS_PATH="/var/lib/plexmediaserver/Library/Application Support/Plex Media Server/Cache/va-dri-linux-x86_64" NEOReadDebugKeys=1 OCL_ICD_VENDORS="/var/lib/plexmediaserver/Library/Application Support/Plex Media Server/Cache/cl-icds-linux-x86_64" X_PLEX_TOKEN=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx cl_cache_dir="/var/lib/plexmediaserver/Library/Application Support/Plex Media Server/Cache/Shaders/icr-3e693db9c71d966fa83dfabd-linux-x86_64/" "/usr/lib/plexmediaserver/Plex Transcoder" "-codec:#0x01" hevc "-hwaccel:#0x01" vaapi "-hwaccel_fallback_threshold:#0x01" 10 "-hwaccel_output_format:#0x01" vaapi "-hwaccel_device:#0x01" vaapi "-codec:#0x02" aac -ss 30 -analyzeduration 20000000 -probesize 20000000 -i '/data/videos/zz-Vidéos-Tests/Sony Swordsmith HDR UHD 4K Demo.mp4' -filter_complex "[0:#0x01]hwupload[0];[0]scale_vaapi=w=1920:h=1080:format=p010[1];[1]hwmap=derive_device=opencl[2];[2]tonemap_opencl=tonemap=mobius:format=nv12:m=bt709:p=bt709:r=tv[3];[3]hwmap=derive_device=vaapi:reverse=1[4];[4]hwupload[5]" -map "[5]" -codec:0 h264_vaapi -b:0 6974k -maxrate:0 9299k -bufsize:0 18598k -r:0 59.939999999999998 -force_key_frames:0 "expr:gte(t,n_forced*1)" -filter_complex "[0:#0x02] aresample=async=1:ochl='stereo':rematrix_maxval=0.000000dB:osr=48000[6]" -map "[6]" -metadata:s:1 language=eng -codec:1 aac -b:1 130k -f dash -seg_duration 1 -dash_segment_type mp4 -init_seg_name 'init-stream$RepresentationID$.m4s' -media_seg_name 'chunk-stream$RepresentationID$-$Number%05d$.m4s' -window_size 5 -delete_removed false -skip_to_segment 31 -time_delta 0.0625 -manifest_name "http://127.0.0.1:32400/video/:/transcode/session/afr7629imvh3hph72qeh2j39/8831ad41-48fa-4fd8-8333-a42963879bd9/manifest?X-Plex-Http-Pipeline=infinite" -avoid_negative_ts disabled -map_metadata -1 -map_chapters -1 dash -start_at_zero -copyts -init_hw_device vaapi=vaapi:/dev/dri/renderD128,driver=iHD -filter_hw_device vaapi -y -nostats -loglevel quiet -loglevel_plex error -progressurl http://127.0.0.1:32400/video/:/transcode/session/afr7629imvh3hph72qeh2j39/8831ad41-48fa-4fd8-8333-a42963879bd9/progress
Jul 01, 2024 23:21:38.936 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39/JobRunner] In directory: "/tmp/Transcode/Sessions/plex-transcode-afr7629imvh3hph72qeh2j39-8831ad41-48fa-4fd8-8333-a42963879bd9"
Jul 01, 2024 23:21:38.936 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39/JobRunner] Jobs: Starting child process with pid 1304
Jul 01, 2024 23:21:38.936 [128682190449464] DEBUG - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Started session successfully: afr7629imvh3hph72qeh2j39

If I interpret correctly your words and the log, the ToneMapping worked fine.

But as I was very tired, I may have missed something in the settings... I don't remember if I checked or not the checkbox...
I'll redo a transcode in the morning (here in France).

And I notice that I get an error for the /transcode directory...
I may go inside my LXC to see what happen.
@tteck Do your script create such a folder?

[ChuckPa]

DUDE!!!!

Jul 01, 2024 23:21:38.936 [128682190449464] ERROR - [Req#5691/Transcode/afr7629imvh3hph72qeh2j39] Error creating directory "/transcode": Permission denied

you have a configuration error (settings - server - transcoder - show advanced)

[chenks]

When LXC is unprivileged, SMB/NFS mount inside it can't work.
You have to set it ti privileged to have smb/NFS mount.

clearly not, my plex LXC is unprivileged.
i have my NAS share mounted (using cifs) in the pve fstab, then mounted within the LXC via the conf file (as mentioned above).

the mounted share works as expected within the LXC.

are you saying that, with a privileged LXC you can mount the share directly in the LXC without going via the fstab on the PVE?

[MilesTEG1]

clearly not, my plex LXC is unprivileged.
i have my NAS share mounted (using cifs) in the pve fstab, then mounted within the LXC via the conf file (as mentioned above).

the mounted share works as expected within the LXC.

You're mistaken, I don't mount my smb share on the pve host... but I mount them inside the lxc, and doing so it need a privileged container.

are you saying that, with a privileged LXC you can mount the share directly in the LXC without going via the fstab on the PVE?
Yes, that's what I'm trying to explain 😁 and what I said before.
The smb mount are fine inside the lxc, not PVE host.

[chenks]

You're mistaken, I don't mount my smb share on the pve host... but I mount them inside the lxc, and doing so it need a privileged container.

i don't think it's worth using privileged containers just to be able to do that.
much safer to use unprivileged even if that means not being able to use snapshots.

[MilesTEG1]

You're mistaken, I don't mount my smb share on the pve host... but I mount them inside the lxc, and doing so it need a privileged container.

i don't think it's worth using privileged containers just to be able to do that. much safer to use unprivileged even if that means not being able to use snapshots.

That's a good point (of view).
But I don't think mounting a SMB share on the PVE host is a good idea too.
I think, there is pros/cons for both.

Just for my information, if I mounted my SMB share on the pve's host, and pass it through to the LXC, where are accessible the data? Can it be changed?
I ask this because my docker installation mounted my medias inside /data/movies, /data/series, ... folders, which are used for my plex libraries.
and when I switched to my actual LXC installation I wanted to keep the data and all infos from my medias and I didn't want to do a re-scan of my files.

[chenks]

Just for my information, if I mounted my SMB share on the pve's host, and pass it through to the LXC, where are accessible the data?

not sure what you mean by that question.
the data is where the data is.

the mount point is no different and the LXC or docker doesn't know any different.

it's the line you add to the LXC conf file that creates the mount point within the LXC.

[ChuckPa]

The default transcode directory is in Cache/Transcode

Transcoder temp directories can never be placed on the network (inotify does not work - needed for Dolby)
(SMB mounts don't work either .. even if local)
map the directory lxc config device add disk.......

maybe best to bring this to the forum and get input from others

[MilesTEG1]

The default transcode directory is in Cache/Transcode

Transcoder temp directories can never be placed on the network (inotify does not work - needed for Dolby) (SMB mounts don't work either .. even if local) map the directory lxc config device add disk.......

maybe best to bring this to the forum and get input from others

I know that :) And my transcode directory isn't on the network.

[Nodiink]

Hi all,
I have the same problem described in this post, at every Proxmox reboot I need to manually change the N of /dev/dri/cardN, switching from 0 to 1 or viceversa, in the resources tab of Proxmox.
The container is for Emby, but I suppose the problem can be on the same origin.
HW is an i5 7500T on a HP minipc.
Here in response I read about problems with transcoding or tone-mapping, in my case everything works fine.

[MilesTEG1]

Hello 👋🏻
It seems that an update of the kernel solved my issue because since I reported this one, I've done several updates of the proxmox host's debian, and this problem seems to not appear again.

[ChuckPa]

I believe I found an answer to this.

1. I just purchased a Dragon Canyon (i9-12900)
2. Ubuntu 22.04.4 HWE (6.5 kernel)
3. There are no issues.

CONVERSELY

1. Xeon E5-2690 v4
2. 5.15 kernel
3. Ubuntu 22.04.4 LTS
4. Nvidia P2200
5. Plex cannot hardware transcode + tonemap + HW burn subtitles.

ROOT CAUSE:

1. Motherboard has a ASPEED IPMI VGA-only display adapter.
2. Kernel INCORRECTLY puts the ASPEED (AST driver) in /dev/dri as card0
3. When P2200 enumerates, it enumerates as card1 and renderD128 (next available nodes)

SOLUTION: (more of a workaround)

Install the Ubuntu HWE kernel.
Xeon now has the 6.5 kernel (soon to be 6.8)
HW transcoding and tonemapping (through the Nvidia P2200) work as expected.
card0, card1, renderD128 still present
NO IDEA why it works but it does

NOTE:
I cannot blacklist the AST driver as that would disable the IPMI. (why I have the IPMI)

[MilesTEG1]

@ChuckPa
You add a discrete nvidia GPU, right?

In my issue, I just have the embedded iGPU (in the Intel CPU).

Your workaround is interesting but not directly related to this issue I guess.

[ChuckPa]

The workaround is because the Linux kernel is broken.

This is Ubuntu 22.04.4 LTS HWE kernel.
There is no discrete GPU here; iGPU only.

[chuck@lizum ~.2010]$ cat /proc/cpuinfo | grep 'model name' | uniq
model name	: 12th Gen Intel(R) Core(TM) i9-12900
[chuck@lizum ~.2011]$ ls -la /dev/dri
total 0
drwxr-xr-x   3 root root        100 Aug 11 21:50 ./
drwxr-xr-x  22 root root       5740 Aug 11 21:50 ../
drwxr-xr-x   2 root root         80 Aug 12 15:06 by-path/
crw-rw----+  1 root render 226,   0 Aug 12 15:06 card0
crw-rw----+  1 root render 226, 128 Aug 11 21:50 renderD128
[chuck@lizum ~.2012]$ uname -a
Linux lizum 6.5.0-45-generic #45~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Mon Jul 15 16:40:02 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
[chuck@lizum ~.2013]$ 

The Xeon box, because it has no iGPU, requires I add a discrete (P2200)

Here is where you see the kernel create the failure as it enumerates the devices into the filesystem.
( The IPMI VGA adapter has no business in /dev/dri )

[chuck@glockner ~.2000]$ 
[chuck@glockner ~.2000]$ cat /proc/cpuinfo | grep 'model name' | uniq
model name	: Intel(R) Xeon(R) CPU E5-2690 v4 @ 2.60GHz
[chuck@glockner ~.2001]$ uname -a
Linux glockner 6.5.0-45-generic #45~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Mon Jul 15 16:40:02 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
[chuck@glockner ~.2002]$ ls -la /dev/dri
total 0
drwxr-xr-x  3 root root        120 Aug 10 13:50 ./
drwxr-xr-x 20 root root       5960 Aug 10 13:51 ../
drwxr-xr-x  2 root root        100 Aug 10 13:50 by-path/
crw-rw----  1 root render 226,   0 Aug 10 13:50 card0
crw-rw----  1 root render 226,   1 Aug 10 13:50 card1
crw-rw----  1 root render 226, 128 Aug 10 13:50 renderD128
[chuck@glockner ~.2003]$ ls -la /dev/dri/by-path/
total 0
drwxr-xr-x 2 root root 100 Aug 10 13:50 ./
drwxr-xr-x 3 root root 120 Aug 10 13:50 ../
lrwxrwxrwx 1 root root   8 Aug 10 13:50 pci-0000:07:00.0-card -> ../card1
lrwxrwxrwx 1 root root  13 Aug 10 13:50 pci-0000:07:00.0-render -> ../renderD128
lrwxrwxrwx 1 root root   8 Aug 10 13:50 pci-0000:0d:00.0-card -> ../card0
[chuck@glockner ~.2004]$ 

What do you show in the by-path subdirectory ?
(by-path shows you the PCI bus addresses of the display adapters which were enumerated)

EDIT: Here you can see the inode creation sequence and where it's broken. 'card0' does not belong here.
card1 and renderD128 should be consecutively numbered as they are but the names should be card0 / renderD128

[chuck@glockner ~.2005]$ ls -lani /dev/dri
total 0
 452 drwxr-xr-x  3 0   0      120 Aug 10 13:50 ./
   1 drwxr-xr-x 20 0   0     5960 Aug 10 13:51 ../
1240 drwxr-xr-x  2 0   0      100 Aug 10 13:50 by-path/
 453 crw-rw----  1 0 109 226,   0 Aug 10 13:50 card0
1058 crw-rw----  1 0 109 226,   1 Aug 10 13:50 card1
1057 crw-rw----  1 0 109 226, 128 Aug 10 13:50 renderD128
[chuck@glockner ~.2006]$ 

It's ALSO possible, with me running the new transcoder code , which has the new ICR in it, that ICR is capable of dealing with this.

We have discussed this problem internally and know of a way to workaround the screwed up / broken kernel naming.
It's not pretty but it can be made to work.

Ask your distro folks (ProxMox) what they can do about the kernel & driver initialization enumeration sequence.
This isn't a Plex problem. We've already confirmed the kernel and, very likely the base video driver for the display, are involved with the problem. (I think it's the base display driver being too aggressive / insufficiently discrete) as it initializes. My machine was fine until the display drivers (not Nvidia) updated.

[Unspec7]

@MilesTEG1 Not sure if this helps, and I don't use plex, but I noticed that when I have a monitor connected to my host on boot, /dev/dri shows card1, and when I don't /dev/dri shows card0. Not sure if you had a monitor plugged in or not.

[MilesTEG1]

@Unspec7
Oh my !
That's maybe the cause!! It's possible that I had the hdmi pluged in my screen... For now, the screen is plugged in another computer ^^
I'll check this out when I have time ^^

[wheelsandcogs]

I think this is definitely related to connected displays - I'm seeing the same thing on an Intel N305 iGPU box when connecting via HDMI.

When I boot it with no display attached, it has /dev/dri/card0. If I have HDMI attached, it's always /dev/dri/card1 (and there is no card0).

[Unspec7]

I think this is definitely related to connected displays - I'm seeing the same thing on an Intel N305 iGPU box when connecting via HDMI.

When I boot it with no display attached, it has /dev/dri/card0. If I have HDMI attached, it's always /dev/dri/card1 (and there is no card0).

I wonder if it's behavior specific to Debian or Linux in general.

[ChuckPa]

It's in the kernel. (been doing research) - simpledrm and simpleframebuffer

Notice the redundancy

[chuck@lizum module.2005]$ sudo lshw -c display
  *-display                 
       description: VGA compatible controller
       product: NVIDIA Corporation
       vendor: NVIDIA Corporation
       physical id: 0
       bus info: pci@0000:01:00.0
       logical name: /dev/fb0
       version: a1
       width: 64 bits
       clock: 33MHz
       capabilities: pm msi pciexpress vga_controller bus_master cap_list rom fb
       configuration: depth=32 driver=nvidia latency=0 mode=3840x2160 visual=truecolor xres=3840 yres=2160
       resources: iomemory:600-5ff iomemory:640-63f irq:275 memory:83000000-83ffffff memory:6000000000-63ffffffff memory:6400000000-6401ffffff ioport:3000(size=128) memory:84000000-8407ffff
  *-display
       description: Display controller
       product: AlderLake-S GT1
       vendor: Intel Corporation
       physical id: 2
       bus info: pci@0000:00:02.0
       version: 0c
       width: 64 bits
       clock: 33MHz
       capabilities: pciexpress msi pm bus_master cap_list
       configuration: driver=i915 latency=0
       resources: iomemory:640-63f iomemory:400-3ff irq:273 memory:645b000000-645bffffff memory:4000000000-400fffffff ioport:4000(size=64) memory:4010000000-4016ffffff memory:4020000000-40ffffffff
  *-graphics
       product: simpledrmdrmfb
       physical id: 1
       logical name: /dev/fb0
       capabilities: fb
       configuration: depth=32 resolution=3840,2160
[chuck@lizum module.2006]$

The result is card0 gets created for the simpledrm driver (which is not used -- but will be with wayland drm -- upcoming)

When the CPU wants to enumerate QSV, we get card1 and renderD128. The nvidia then enumerates and gets card2 and renderD129

Verified here.

[chuck@lizum module.2006]$ ls -la /dev/dri
total 0
drwxr-xr-x   3 root root        160 Sep  8 22:12 ./
drwxr-xr-x  22 root root       5620 Sep  9 16:04 ../
drwxr-xr-x   2 root root        140 Sep  9 12:01 by-path/
crw-rw----+  1 root render 226,   0 Sep  8 22:12 card0
crw-rw----+  1 root render 226,   1 Sep  8 22:12 card1
crw-rw----+  1 root render 226,   2 Sep  9 12:01 card2
crw-rw----+  1 root render 226, 128 Sep  8 22:12 renderD128
crw-rw----+  1 root render 226, 129 Sep  8 22:12 renderD129
[chuck@lizum module.2007]$ ls -la /dev/dri/by-path/
total 0
drwxr-xr-x 2 root root 140 Sep  9 12:01 ./
drwxr-xr-x 3 root root 160 Sep  8 22:12 ../
lrwxrwxrwx 1 root root   8 Sep  8 22:12 pci-0000:00:02.0-card -> ../card1
lrwxrwxrwx 1 root root  13 Sep  8 22:12 pci-0000:00:02.0-render -> ../renderD128
lrwxrwxrwx 1 root root   8 Sep  9 12:01 pci-0000:01:00.0-card -> ../card2
lrwxrwxrwx 1 root root  13 Sep  8 22:12 pci-0000:01:00.0-render -> ../renderD129
lrwxrwxrwx 1 root root   8 Sep  8 22:12 platform-simple-framebuffer.0-card -> ../card0
[chuck@lizum module.2008]$ 

[sirjmann92]

This issue is not related to either Plex or Proxmox. I have the same issue on a mini PC running Ubuntu 24.10, and it impacts both Plex and Jellyfin. I have several mini PCs in a server rack and I have an HDMI KVM switch. Sometimes when I reboot the 24.10 machine, I get /dev/dri/card0, sometimes I get card1. I've tried to force it to card0 by making sure my HDMI KVM switch was on the input for that machine so it detected a monitor connected, but I still got card1 after a reboot. I don't know if maybe I need to leave it on a different input for longer before I reboot or what, but it seems random to me, rather than related to a monitor being connect, at least in my case.

I did not have this issue with Ubuntu 24.04, so I believe it was something introduced with the new kernal (6.11?) or it could be related to something else that was also updated between 24.04 and 24.10....

Is it possible to write a script that will always symlink card1 to card0 if card0 doesn't exist? Then create a service for it so it runs at boot? Obviously it would be a hacky workaround, if it even worked, but until the root cause is identified and issue corrected...

[tteck]

#3096

[sirjmann92]

Here's how I addressed this (with udev rules):

Create a script to symlink to a new dri location, regardless of /dev/dri card#

#!/bin/bash

# Check if /dev/dri/card1 exists
if [ -e /dev/dri/card1 ]; then
  # Create or update the symlink to point to card1
  ln -sf /dev/dri/card1 /dev/dri/intel-igpu
else
  # Create or update the symlink to point to card0
  ln -sf /dev/dri/card0 /dev/dri/intel-igpu
fi

Make the script executable:
chmod +x dri-symlink.sh

Create a new udev rule:
sudo nano /etc/udev/rules.d/99-intel-dri-symlink.rules

Reference the script created earlier in the udev rule by inserting this snippet into the new rule file:

# Trigger the script when the GPU device is added or changed
KERNEL=="card[01]", SUBSYSTEM=="drm", RUN+="/path/to/dri-symlink.sh"

Save and exit nano

Reload the udev rules (or reboot if this doesn't work)
sudo udevadm control --reload-rules && sudo udevadm trigger

Verify the symlink was created:
ls -l /dev/dri

Update docker containers (Jellyfin, Plex, etc.) to point the new symlink on the host to card0 in the container:
/dev/dri/intel-igpu:/dev/dri/card0

Start the container to test. If it works, restart to host to test persistence.

[tteck]

Update docker containers (Jellyfin, Plex, etc.)

The Jellyfin and Plex scripts are not Docker-based.

[sirjmann92]

You didn't read my post, I said I wasn't using Proxmox, I'm using Docker on Ubuntu. This is not a Plex or Proxmox issue... My proposed solution could be adapted to Proxmox, I was just posting here in case others came across the same issue and looking for a resolution (that's how I found it).