AMD Posts Open-Source Linux Driver Support For "NAVI 14" GPU

[L_A_G]

AMD was kind of cutting it close getting Navi 10 support into the DRM-Next repo and Mesa. My initial assessment was that they were just playing their cards close to their chest to avoid spilling out product details ahead of the official announcement, but this would suggest that there was a legitimate non-marketing reason for that and I was wrong in my initial assessment.

[asdfgh]

Is there a big ticket for this ? I didn't see anything in bugzilla for Mudrunner

Nope. Maybe when I'm in mood for bug hunting I'll prepare it with all required debug information. At least I know how to avoid it.

but you mentioned "first bug report was on 2018-02-26". VMC page fault errors are not a specific bug, they just mean that something in the application or driver accessed memory that it didn't own, and that could have dozens or hundreds of different root causes.

vega_crasher from https://bugs.freedesktop.org/show_bug.cgi?id=105251#c52 reproducibly crashes my APU. It looks like it's based on learnopengl.org tutorials, a very simple and correct application.

Did you or your system updater change any files yesterday between working and not working ? I wouldn't expect the system to just start generating errors on its own.

See my comment in https://bugs.freedesktop.org/show_bug.cgi?id=105251#c69 I have just rewritten immediate mode code to use VBO+VAO. The only differences between vega_crasher and my code is that I use gl2.1 and VAO if available, it uses gl3.3 and I use GL_TRIANGLE_FAN (4 vertices), it uses GL_TRIANGLES (6 vertices).

Reading through the bug ticket it appears the primary problem (VM faults when running cemu under Wine) was fixed in cemu upstream, and what is left appears to be a collection of unrelated problems that only have VM faults in common.

Was it? Or cemu dev just added workarounds? There is no notification from amdgpu or mesa devs that they have figured out the root cause of the issue.

I see that Vega or drivers for it are unable to handle tiny vertex buffers. Funny thing that immediate mode still works.

[slowee]

Frankly speaking I'm not an owner of any "big" vega (56, 64 or 7). I have a 2400G APU and it's terribly unstable. Usually when I play Mudrunner, but since yesterday I've being hit by "[gfxhub] VMC page fault" errors (the first bug report was on 2018-02-26).

From time to time I'm thinking about buying a new more or less powerful videocard, but maybe it's better to wait a year more to see what Intel will propose.

I'm an owner of the same hw, and i'll never get it away.

Until December last years i was probably in the same freakin situation like you - freezes over crashes, no matter what i did, nothing could make this damn thing stable.
Tried several mesa/libdrm/kernel/UEFI combinations and options - waste of time.

First as ASUS published UEFI update 4207 for the mb, everything changed.

Now it runs as it should in May 2018 (where i bought it) - no freezes, no crashes.

From time to time i use an old MS Windows game with wine to do some Tests, and this is causing issues like yours (VMC page fault or ring gfx0 sequence mismatch). Bad software - bad mood.

[entropy]

vega_crasher from https://bugs.freedesktop.org/show_bug.cgi?id=105251#c52 reproducibly crashes my APU. It looks like it's based on learnopengl.org tutorials, a very simple and correct application.

See my comment in https://bugs.freedesktop.org/show_bug.cgi?id=105251#c69 I have just rewritten immediate mode code to use VBO+VAO. The only differences between vega_crasher and my code is that I use gl2.1 and VAO if available, it uses gl3.3 and I use GL_TRIANGLE_FAN (4 vertices), it uses GL_TRIANGLES (6 vertices).

Hope the devs will get back to this.
If vega_crasher still hangs your hardware, this should be an ideal reproducible bug report, no?

[entropy]

Plot twist: Sample program was incorrect.
They're working on a fix because this should not hang the GPU.

Great!