Hacker News Reader

Comment by eqvinox | Hacker News Reader

> But I am stupid enough to use an nvidia card, which means that I have no Wayland. I am eating shit due to the drivers every few months.

Funny you would say this; I'm on AMD (Ryzen 6950HS IGD) and I'm eating shit due to the (FOSS) drivers every few days… to be fair I have two choices of distinct shit with X11 and Wayland, but neither is actually stable.

And none of this affects gaming, just the desktop in general :/. The RX 6950 in the desktop is in fact working without issues (same driver stack). I suspect it's some memory management problem with the CPU/GPU sharing on the 6950HS. But I have no idea how to track it down…

That said, FOSS GPU support has come a long way.

View on HN

Laptop APU https://www.amd.com/en/products/apu/amd-ryzen-9-pro-6950hs

You don't describe the problems with anywhere near enough detail to guess at what the underlying issue might be.

https://wiki.archlinux.org/title/AMDGPU << Has a troubleshooting section that might help.

> You don't describe the problems with anywhere near enough detail to guess at what the underlying issue might be.

That's because this is the HN comment section, not the f.d.o bug tracker.

Since you seem to care, what I'm seeing is likely #2220, though the real problem for me is that dmesg output is not particularly helpful on its own and it's the entire desktop that crashes (as opposed to a single application that you could set Mesa/amdgpu debug flags on.) The fact that it "only" crashes about once per week complicates things further; aimlessly enabling debug options would probably result in getting buried in logs.

  [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring sdma0 timeout, signaled seq=24191109, emitted seq=24191111
  [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process  pid 0 thread  pid 0

This is really kind of a debugging worst case. There's pretty much no useful information here.

    echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level

Or try setting the kernel boot parameter

    amdgpu.ppfeaturemask=0xfffd3fff

Thanks but I wasn't asking for suggestions, forcing to a high power mode is not helpful on a laptop, and various different values for ppfeaturemask have been discussed in the fd.o bug (none of which helped tackle the issue).

I don't expect it to be the cause, but who knows: does your GPU happen to be sitted in the bottom-most PCIe slot of the motherboard ? I encountered weird stuff with the PCIe lines connected to the motherboard chipset (instead of those connected directly to the CPU) on several B550. Some PCIe packets would apparently be garbled/lost when under load, resulting in the driver for the associated peripheral timing out.

This is an APU in a laptop ("Ryzen 6950HS IGD")

Reply on news.ycombinator.com