Amdgpu-install nuked my system

Ask Fedora
, , , ,
1

Hello lads! Need your help. So i have MSI Delta 15 laptop with RX6700M gpu running Fedora 38 kernel 6.5.9-200 . I’ve been using it for more than a year and i am comfortable with terminal. Two days ago i installed Blender but encountered an issue: hip rendering isn’t working on opensource drivers. I downloaded amdgpu-install script from official website (rhel 9.1), followed all the instructions but at the end i got a build error that said that it couldn’t build for my running kernel and spitted stacktrace. It also said that something was up with dkms. I run the command again sudo amdgpu-install --usecase=workstation --vulkan=pro --opencl=rocr and got:

Last metadata expiration check: 0:05:05 ago on Tue 28 Nov 2023 10:56:03.
Package amdgpu-pro-23.30-1684442.el9.x86_64 is already installed.
Package amdgpu-dkms-1:6.2.4.50702-1683306.el9.noarch is already installed.
Package rocm-opencl-runtime-5.7.2.50702-110.el9.x86_64 is already installed.
Package vulkan-amdgpu-pro-23.30-1684442.el9.x86_64 is already installed.
Dependencies resolved.
Nothing to do.
Complete!
WARNING: amdgpu dkms failed for running kernel

sudo dkms status gave me: amdgpu/6.2.4-1683306.el9: added

then i run: sudo dkms install -m amdgpu -v 6.2.4-1683306.el9
Building module:
Cleaning build area…(bad exit status: 2)
make -j16 KERNELRELEASE=6.5.12-200.fc38.x86_64 TTM_NAME=amdttm SCHED_NAME=amd-sched -C /lib/modules/6.5.12-200.fc38.x86_64/build M=/var/lib/dkms/amdgpu/6.2.4-1683306.el9/build…(bad exit status: 2)
Error! Bad return status for module build on kernel: 6.5.12-200.fc38.x86_64 (x86_64)
Consult /var/lib/dkms/amdgpu/6.2.4-1683306.el9/build/make.log for more information.

I had enough for that day so i rebooted the system and got an error:
[TIME] Timed out waiting for device
[DEPEND] Dependency failed for

The next day i chrooted from live usb into my system and in an attempt to fix the issue i tried to install older kernel headers and somehow got kernel 6.2. BTW it didn’t display the new kernel in GRUB.

Today i chrooted in the system again, run dnf update and updated to new 6.6 kernel that DID showed up in GRUB but when booting into it i get:
error: …/…/grub-core/fs/btrfs.c:2120:file ‘/boot/vmlinuz-6.6.2-101.fc38.x86_64’ not found
error: …/…/grub-core/loader/i386/efi/linux.c:258:you need to load the kernel first

I think that is because when chrooting i only mounted /proc /dev and /sys according to this guide: How to: Chroot into a broken system via live CD/ISO or alternate Linux system | TurnKey GNU/Linux

Guys, did i completely bricked my system? I don’t want to reinstall everything from zero… I know i should’ve used something like Timeshift prior. Any suggestions?

2

error: …/…/grub-core/fs/btrfs.c:2120:file ‘/boot/vmlinuz-6.6.2-101.fc38.x86_64’ not found

It looks like grub is just saying that it cannot find the kernel.

WARNING: amdgpu dkms failed for running kernel

That dkms error might be what is preventing your kernels from installing properly. Assuming you don’t actually need dkms for anything, touch /etc/dkms/no-autoinstall should disable it globally. (Or if you can get it completely uninstalled, that would be better yet. Having that file under /etc/dkms shouldn’t hurt either way.)

3 
Package amdgpu-pro-23.30-1684442.el9.x86_64 is already installed.
Package amdgpu-dkms-1:6.2.4.50702-1683306.el9.noarch is already installed.
Package rocm-opencl-runtime-5.7.2.50702-110.el9.x86_64 is already installed.
Package vulkan-amdgpu-pro-23.30-1684442.el9.x86_64 is already installed.

Did you install those from the epel repo or am I reading that wrong.
4

As I understand it the amdgpu-pro package is discouraged as it causes problems with fedora and users are encouraged to use the default provided drivers for amd GPUs. Installing drivers from a 3rd party repo, even a redhat repo can often cause problems.

My recommendation – for what it is worth – is to remove all those driver packages that are not part of the fedora distro.

1 Like
5

How do i install regular drivers then? When chrooting i uninstalled proprietary drivers with amdgpu-install --uninstall which removed those packages:
amdgpu-dkms
noarch 1:6.2.4.50702-1683306.el9 @amdgpu 421 M
amdgpu-dkms-firmware
noarch 1:6.2.4.50702-1683306.el9 @amdgpu 76 M

Then i run dnf install amdgpu and i got this:
Installing:
amdgpu x86_64 1:5.7.50702-1683306.el9 amdgpu 6.4 k
Installing dependencies:
amdgpu-core noarch 1:5.7.50702-1683306.el9 amdgpu 8.7 k
amdgpu-dkms noarch 1:6.2.4.50702-1683306.el9 amdgpu 9.3 M
amdgpu-dkms-firmware noarch 1:6.2.4.50702-1683306.el9 amdgpu 11 M

I think the repository should’ve been ‘fedora’ or something because i gon a bunch of errors:

EFI variables are not supported on this system
Loading new amdgpu-6.2.4-1683306.el9 DKMS files…
It is likely that 6.2.9-300.fc38.x86_64 belongs to a chroot’s host
Building for 6.2.7-200.fc37.x86_64, 6.2.8-200.fc37.x86_64, 6.2.9-200.fc37.x86_64, 6.2.11-200.fc37.x86_64, 6.2.11-300.fc38.x86_64, 6.2.13-300.fc38.x86_64, 6.2.14-300.fc38.x86_64, 6.2.15-300.fc38.x86_64, 6.3.4-201.fc38.x86_64, 6.3.5-200.fc38.x86_64, 6.3.7-200.fc38.x86_64, 6.3.8-200.fc38.x86_64, 6.3.11-200.fc38.x86_64, 6.3.12-200.fc38.x86_64, 6.5.10-200.fc38.x86_64, 6.5.12-200.fc38.x86_64 and 6.6.2-101.fc38.x86_64
Module build for kernel 6.2.7-200.fc37.x86_64 was skipped since the
kernel headers for this kernel do not seem to be installed.

Error! Bad return status for module build on kernel: 6.5.10-200.fc38.x86_64 (x86_64)

Error in POSTIN scriptlet in rpm package amdgpu-dkms
Running scriptlet: amdgpu-core-1:5.7.50702-1683306.el9.noarch 3/4
ERROR: This package can only be installed on EL9.
error: %prein(amdgpu-core-1:5.7.50702-1683306.el9.noarch) scriptlet failed, exit status 1

Error in PREIN scriptlet in rpm package amdgpu-core
Installing : amdgpu-1:5.7.50702-1683306.el9.x86_64 4/4
error: amdgpu-core-1:5.7.50702-1683306.el9.noarch: install failed

Running scriptlet: amdgpu-1:5.7.50702-1683306.el9.x86_64 4/4
Verifying : amdgpu-1:5.7.50702-1683306.el9.x86_64 1/4
Verifying : amdgpu-core-1:5.7.50702-1683306.el9.noarch 2/4
Verifying : amdgpu-dkms-1:6.2.4.50702-1683306.el9.noarch 3/4
Verifying : amdgpu-dkms-firmware-1:6.2.4.50702-1683306.el9.no 4/4

Installed:
amdgpu-1:5.7.50702-1683306.el9.x86_64
amdgpu-dkms-1:6.2.4.50702-1683306.el9.noarch
amdgpu-dkms-firmware-1:6.2.4.50702-1683306.el9.noarch
Failed:
amdgpu-core-1:5.7.50702-1683306.el9.noarch

Error: Transaction failed

Do i need to install custom headers? Even tho i run dnf update and updated the kernel to 6.6, uname -a still gives me:
Linux localhost-live 6.2.9-300.fc38.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Mar 30 22:32:58 UTC 2023 x86_64 GNU/Linux

And grub doesn’t show the 6.2 version.

And should i really use fedora at this point? The latest amdgpu drivers are for rhel 9.2 which seems to run kernel 5.14 instead of 6.6. I really want to have hip rendering in Blener ( that’s why it all started after all ).

6

I’ve been told that the amdgpu driver that comes included with Fedora Linux is the latest version. You don’t need to install anything.

7

Disable the epel repo before doing so. That is trying to reinstall packages from epel and that may cause conflicts. Especially since the fedora version of blender is working for most users with direct install from fedora.

8

This is correct: everything needed for typical gaming is already provided by official Fedora pkgs. For a while now, Fedora has had everything you (typically) need. My RDNA3 card is very fast for games.

amdgpu-install will not even run on Fedora without manual changes to the scripts, and, AFAIK, there hasn’t been a successful report of using amdgpu-install on Fedora for about 10 years. I don’t think anyone knows how to get it to work. Fedora is not supported by AMD, and that installer really shows how much they DON’T support Fedora. That installer is for distributions that don’t have any kernel support for AMD, but that’s basically not any distribution because they AMDGPU is in the mainline kernel. So, the only thing you can get out of that installer is AMDGPU-PRO, which some people want. But, usually not Fedora folks because Fedora is strictly OSS.

That’s not to say AMDGPU (non PRO) is perfect. There is a driver problem with OpenCL workloads and RDNA3 cards, and you don’t need OpenCL for games if that’s what you want. This is one of the things I assume would work better with -PRO, but I’ll never know because I don’t want to convert my whole system to another distribution.

The other thing that is missing in Fedora is full HIP support. But, you don’t need that for games either. These two issues (OpenCL & HIP) have to do with how many features go into the OSS driver stack for Linux, which typically has been enough for lots of people. My RDNA2 card runs OpenCL fine, for example, with the exact same drivers. This has been an unfortunate year without the typical OSS support, likely related to the fact that RDNA3 is just one year old, although I think just 6 months lag would have been quite long. Still waiting.

1 Like
9

Blockquote
Especially since the fedora version of blender is working for most users with direct install from fedora.

Okay, so, I think Blender has to do with the HIP support I mentioned. Sounds like HIP is getting better on Fedora too. Great to hear.

10

I do game on my system and they are running great, but I want the HIP rendering that is only available with proprietary CL drivers. Do you suggest switching my distro to ubuntu which IS supported?

11

Well, I don’t use Blender, so I suggest you try to find people who use it on Fedora. @computersavvy helped me before, so I’d trust his statement that HIP is working better now. That sounds promising to me. Have you gone back to the full default setup in Fedora, yet? I’d at least give that a try.

There was someone else who posted about the same thing about 6 months ago, Blender + AMDGPU. I don’t remember what happened, but I would seek out that thread. Team red can feel kind of lonely, sometimes, regardless of what OS you use, but more so on Linux. But, I use AMD because it works on Fedora at all.

If you have another video card, you could try dedictating the AMD card to a VM and install another OS there to test. That’s a lot of work, but so is installing a new OS on hardware, and at least you wouldn’t have to make your whole system unusable while testing.

I’ve been in contact with the maintainer of the ROCm pkg, and that person reported that ROCm-6 is on the horizon. Maybe wait for that? I’m putting all my hopes on that.

12

Oh, if you are sure the OSS stack is never going to get the support you need, then, no, you cannot use Fedora or any of the other officially unsupported distributions, AFAIK. So that leaves you with, what?, Ubuntu/Debian, SuSE, RedHat ?

13

Are you referring to Open Source Software?
If so then it would appear obvious, since the OP claims what he wants is supported on ubuntu, that your reference may be off a bit. Different distros focus on different paths with the software they include (and versions of that software) so your comment would be inaccurate. It would seem that OSS does support what he asks about but may not be available at present in fedora for different reasons.

14

Excuse me. . .

We’ve needed the drivers for a long time, but about 2/3yrs ago people started to unpack them and build them on their own GE (who’s known for Proton packages) also did this to help the community who use Blender and Da Vinci Resolve. it’s been working for about 2yrs.

15

Wow! Fantastic! That’s amazing news from the GloriousEggroll folks!

I don’t know why there’s no activity history, but it looks to me like that began about 2 years ago.

Sounds like @internet-user95 is getting an early present. Fedora Discussion to the rescue!

1 Like
16

I don’t want to fish through the forum history, I think this forum was combined some time ago, but I also ran into this and had post about it. No need to look to the past now.

Here’s the latest :

and here is some helpful tips for people down the line/future that might come in handy:

https://radeon-pro.github.io/RadeonProRenderDocs/en/plugins/blender/installation.html