Software > Operating Systems and Porting
Debian 12 status?
pocock:
I took the 6.7.12-1 backport for bookworm and applied the 4k page size patch.
It has the same problem with amdgpu that I saw using the 6.1 kernel with 4k patch. Errors about "Not enough memory for command submission!" and eventually Firefox and the whole desktop freeze.
I pushed each of my branches for anybody else who wants to try it. Maybe it is fine with other GPUs
This is the 4k patch against the 6.7.12-1 kernel on the branch for unstable and testing
The same patch backported for the 6.7.12-1 kernel on bookworm-backports
The 6.1 standard kernel branch for bookworm with the 4k page size patch
There are various other web sites discussing the error message. Some of them suggest that the amdgpu firmware needs to be updated.
One of them points to a new version of the amdgpu firmware that appeared in the upstream kernel tree in June 2024
At present, the Debian firmware package still has older amdgpu firmware from 2023 - you can see the current versions here, even unstable hasn't been updated since June 2023.
Here is the packaging repository on Debian Git (Salsa), we can't clearly see the actual versions of the amdgpu firmware files
MauryG5:
Sorry but why are you using the 4K page size? AMDGPU drivers have long since eliminated the bug that occurred at the time for the default page sizes used on Power, i.e. 64K. I have long since returned to 64K pages on the Kernel and I have not had any problems for a long time on my Radeon 5700XT. Why do you have to use that size? I still use 11.9 as my Debian version, I have not wanted to update yet because I like Gnome 3.38 better as it is more original in its animations and use, compared to the current Gnome 3.4X... When I am forced to update then I will.
Hasturtium:
4KB page sizes are still required for the nouveau driver, and may be useful in bringing Intel's xe up on ppc64le too. The maintainer of Void Linux stood by the practice too out of personal preference and cross-platform compatibility.
pocock:
I put some of my observations in this blog post
The BtrFs issue is particularly annoying if you create a BtrFs using 4k on an external drive or USB stick and you want to attach that drive to a host running the 64k page size.
They made some improvements to BtrFs in more recent kernels but last time I looked at it they still said the combination of 64k page size with 4k BtrFs block size is experimental. Then again, they still tell us the BtrFs itself is experimental.
Porting applications to POWER and porting to 64k are two different problems. Trying to solve them both at the same time in an environment where at least some people are here as hobbyists may be asking too much.
Hasturtium:
--- Quote from: pocock on July 30, 2024, 03:13:41 am ---I put some of my observations in this blog post
The BtrFs issue is particularly annoying if you create a BtrFs using 4k on an external drive or USB stick and you want to attach that drive to a host running the 64k page size.
They made some improvements to BtrFs in more recent kernels but last time I looked at it they still said the combination of 64k page size with 4k BtrFs block size is experimental. Then again, they still tell us the BtrFs itself is experimental.
Porting applications to POWER and porting to 64k are two different problems. Trying to solve them both at the same time in an environment where at least some people are here as hobbyists may be asking too much.
--- End quote ---
Yeah, I'd forgotten about Btrfs, that's potentially a big one. As a self-avowed hobbyist and minimal dabbler in coding, that much is definitely true.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version