AMD Posts Patches Implementing RAS Support For AMDGPU Linux Driver

[phoronix]

Phoronix: AMD Posts Patches Implementing RAS Support For AMDGPU Linux Driver

AMD developers today posted a set of twenty patches implementing RAS support for the AMDGPU Linux kernel diver...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-RAS-Patches-Linux

[boxie]

I kinda hope that this stuff makes it back to Vega 56/64 - while the lockups are occasional riding the padoka ppa, they are annoying

[Buntolo]

Does this mean that a crash will just result in a driver reload, like on Windows, instead of a kernel panic?

[uid313]

Does this mean that a crash will just result in a driver reload, like on Windows, instead of a kernel panic?

The Linux kernel panics with AMD and Nvidia drivers?
I use Intel integrated graphics and my Linux experience is rock solid. I never experience any crashes.

A time out can cause a GPU reset. The Linux kernel is a monolithic kernel, drivers are built in. The problem with the GPU reset is that X can not survive of that. The Amdgpu GPU reset mechanism is for servers.

What about Wayland? Can it survive a GPU reset?

[uid313]

Propably not, wayland can not even survive of daily tasks. Forget wayland, it will never replace X. 10 years of development and very few native wayland applications, makeshift solutions for X and so on.

I believe Wayland already replaced X.Org Server in Fedora.
All of GNOME and KDE are native Wayland applications. Most GTK 3 and Qt 5 applications are native Wayland applications.
Only left now is Firefox (nightly have support), Chromium, GIMP and Electron-based applications (since they derive from Chromium).

[agd5f]

This has nothing to do with GPU reset per se. It's mainly enabling on chip/board reliability features like ECC.

[agd5f]

What about Wayland? Can it survive a GPU reset?

Wayland and X are in the same boat. The display server (and optionally applications themselves) need to enable API specific robustness support to detect when the context is lost and restore their buffer state. E.g., gl_arb_robustness, or the context lost stuff in vulkan. The reason GPU reset works on windows is because Microsoft uses the equivalent DX robustness API in the windows display manager to catch context lost errors and restore its buffers. Even on windows individual applications still need to catch these errors to survive a GPU reset.