[TRACKING] Graphical corruption in Fedora 39 (AMD 3.03 BIOS)

Rei · March 24, 2024, 4:12am

I happened to trigger that graphic corruption today, after I turned off UMA_GAME_OPTIMIZED.

However, switching back and forth TTY doesn’t work for me the key combo I used were ctrl+alt+F3 then ctrl+alt+F2.

Also I’m on Kernel 6.6.10 Pop!_OS, does upgrading to the latest kernel help?

mikeymop · March 24, 2024, 4:21am

Unfortunately I can still experience this on 6.8.1. It actually happens much more often on the newer kernels unless you enable UMA_GAME_OPTIMIZED.

However, overall power consumption is much lower on 6.8.1 so I trade the extra ram for the battery life and continue to test each new kernel.

Rei · March 24, 2024, 4:24am

may I ask what distro are you using?

I bet I’d do that same - enable UMA_GAME_OPTIMIZED

Until now, I don’t have an effective way to trigger the white screen, i.e. I don’t know what will trigger Lol

R1ck · March 24, 2024, 10:00pm

mhh sadly this doen’t work on arch with sway ^^v

mikeymop · March 25, 2024, 3:05am

I’m running the Fedora 40 branch (eg: fc39 dnf system-upgrade to fc40).

baptistemm · March 25, 2024, 7:27am

Hello,

This thread is becoming way too long, is there a place that give the current status and the possible workaround ? in the KB for instance ?
Just to say also under fedora 40 the graphical corruption is get worse.

Arthur_Bols · March 25, 2024, 9:28am

I’m running the Fedora 40 branch (eg: fc39 dnf system-upgrade to fc40).

Don’t do this as a novice! The beta isn’t even out yet.

Just install the kernel manually.

Arthur_Bols · March 25, 2024, 9:36am

There is nothing that solves everything, but there are workarounds.

Kernel >= 6.8 has many improvements, but still has some white screen issues (very weird that it’s worse for you)
Change the BIOS settings from Auto to UMA_Game_Optimized.

This shouldn’t be necessary anymore in 6.8, but if you encounter issues feel free to set it:

Set kernel arg amdgpu.sg_display=0

baptistemm · March 25, 2024, 10:25am

Yes, this is I read a couple of month ago, that motivated me to migrated to Fedora 40 (that I do at time of beta anyway)
That why I explain all cases I encounter screen corruption, because it is now unusable. I have to restart 2 times per day.

I already did, but perhaps it switched back to AUTO, I’ll double-check

Arthur_Bols · March 25, 2024, 11:45am

I’ve created a framework backports repo. It currently only contains the kernel from Fedora 40. Let me know if more packages are required.

It’s updated once a day by downloading the packages from Fedora 40 and uploading them to this repo. Nothing is modified, so I’ve enabled the Fedora gpgkey’s.

USE AT YOUR OWN RISK!

sudo tee /etc/yum.repos.d/framework-backports.repo << "EOF" > /dev/null
[framework-backports]
name=Framework backports
baseurl=https://principis.fedorapeople.org/framework-backports/$releasever/
enabled=1
metadata_expire=7d
repo_gpgcheck=0
type=rpm
gpgcheck=1
gpgkey=file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora-40-$basearch file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora-41-$basearch file:///etc/pki/rpm-gpg/RPM-GPG-KEY-fedora-rawhide-$basearch
EOF

To remove this repo (and installed packages):

sudo rm -f /etc/yum.repos.d/framework-backports.repo
sudo dnf distrosync --refresh
sudo dnf remove "kernel-*fc40*" # THIS SHOULD ONLY REMOVE KERNEL* PACKAGES!!!

Alex_H · March 26, 2024, 3:18pm

I still get the flicker occasionally.

I’m running VMs via libvirt and when resizing or fullscreening them on my external 4k monitor, I occasionally either freeze the VM or my entire system with this kernel error:

[ 2034.021192] gmc_v11_0_process_interrupt: 46 callbacks suppressed
[ 2034.021197] amdgpu 0000:c1:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:3 pasid:32770, for process qemu-system-x86 pid 5607 thread qemu-syste:cs0 pid 5642)
[ 2034.021205] amdgpu 0000:c1:00.0: amdgpu:   in page starting at address 0x0000000018000000 from client 10
[ 2034.021209] amdgpu 0000:c1:00.0: amdgpu: GCVM_L2_PROTECTION_FAULT_STATUS:0x00301430
[ 2034.021211] amdgpu 0000:c1:00.0: amdgpu: 	Faulty UTCL2 client ID: SQC (data) (0xa)
[ 2034.021213] amdgpu 0000:c1:00.0: amdgpu: 	MORE_FAULTS: 0x0
[ 2034.021216] amdgpu 0000:c1:00.0: amdgpu: 	WALKER_ERROR: 0x0
[ 2034.021217] amdgpu 0000:c1:00.0: amdgpu: 	PERMISSION_FAULTS: 0x3
[ 2034.021219] amdgpu 0000:c1:00.0: amdgpu: 	MAPPING_ERROR: 0x0
[ 2034.021221] amdgpu 0000:c1:00.0: amdgpu: 	RW: 0x0
[ 2044.180516] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx_0.0.0 timeout, but soft recovered

CGE · March 27, 2024, 4:00am

In my case, on Fedora branched/40, with 6.8.1-300.fc40, I was seeing rather frequent white screen corruption, always full-screen, and after setting amdgpu.sg_display=0, have not seen it since, though it’s only been around 2 days. I have 64 GB of RAM (2x32) from FW.

As setting amdgpu.sg_display=0 seems like it should be redundent, given the SG disabling code at ≥ 64 GB, I wonder if that check is no longer in the most recent branched kernel for some reason, whether it was an attempt to get SG legitimately working in this case, or something else? I haven’t had a chance to check.

mikeymop · March 28, 2024, 3:30am

Thats interesting as I have 32gb memory and experienced the glitches as well. However, rather than disabling scatter/gather I set UMA_GAME_OPTIMIZED to lock the gpu memory size and this has proven stable as well.

I’m not sure which method is preferred but it definitely seems like a subtle memory leak or overflowed buffer happening around the shared memory region.

Rei · March 29, 2024, 8:32am

I have only used the UMA_GAME_OPTIMIZED method, and never have any glitches when it’s enabled.

Btw, after recent BIOS update, I got more available ram. 29 GB vs 27 GB.

dimitris · March 29, 2024, 5:12pm

3.02 to 3.03 or 3.03 to 3.03b?

Also, after the BIOS update, is UMA_GAME_OPTIMIZED still set?

Rei · March 30, 2024, 9:01am

sorry I should’ve specified… it’s 3.03 to 3.03b

UMA_GAME_OPTIMIZED still set, without it the system is more prone to glitch

dimitris · March 30, 2024, 4:40pm

Thanks. Weird that an EC only update, focused on power/charging, would change available memory…

Rei · April 7, 2024, 8:48am

I disabled UMA_GAME_OPTIMIZED today. Btw its called “Gaming” after 3.05.

I tried to trigger the corruption, like I did last time… open many tabs running YouTube, Totem running 1080p video, and plugging and unplugging my usb-c dock with monitor connected.

I opened 30 tabs in the background and plug/unplug dock, close/open lid, on/off fractional scaling, open/close lid - didn’t trigger white graphical corruption.

I used mine whole day without any issue.

worth noting, I’m on kernel 6.8, bios 3.05. no amdgpu.sg_display=0

Rei · April 7, 2024, 8:52am

So the setting for iGPU memory is as follow:
Auto: Allocate 512MB to FPU if system memory is below 64GB. else 2GB.
Gaming: Allocate 2GB to GPU if system memory is between 8-24GB. 4GB if system memory is 24GB and above.

Is 512MB enough for normal use?

Mario_Limonciello · April 7, 2024, 11:04am

When S/G is on absolutely it is. Once that 512 is exhausted the kernel driver will pull from GTT.