Bug 99292

Summary: GPU hang in High Fidelity
Product: Mesa Reporter: Christoph Haag <haagch>
Component: Drivers/Gallium/radeonsiAssignee: Default DRI bug account <dri-devel>
Status: RESOLVED FIXED QA Contact: Default DRI bug account <dri-devel>
Severity: normal    
Priority: medium    
Version: git   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments: dmesg

Description Christoph Haag 2017-01-05 23:47:43 UTC
Created attachment 128789 [details]
dmesg

RX 480 with llvm 4.0.0svn_r291124, mesa git, Linux 4.10-rc2

It works for some time, but randomly clicking through the UI with the address bar hangs the GPU.

Here is an apitrace (18 Megabyte): https://haagch.frickel.club/files/interface.trace.xz
Comment 1 smoki 2017-01-06 04:39:58 UTC
 This is same trace like in closed bug 99191 that no driver could run normaly.
Comment 2 Christoph Haag 2017-01-06 08:54:45 UTC
Oops, it has the same name, so I overwrote the other one. It really should be a new trace though.
Comment 3 Samuel Pitoiset 2017-01-13 16:28:30 UTC
Confirmed, I can reproduce the hang on my rx 480 with latest mesa/llvm.

It hangs at draw call 680247, looks like an infinite loop in the fragment shader.
Comment 4 Christoph Haag 2017-01-17 13:03:21 UTC
Someone said that with mesa with --enable-debug it runs in an assert before the gpu hang happens, so I tried it to add it to the bug report, but this did not happen for me. I get no assert and the GPU still hangs with --enable-debug.

Just an idea, since High Fidelity has had the other known application side rendering bug, it's possible there is a buggy shader that the compiler doesn't catch.
Comment 5 Christoph Haag 2017-07-06 17:11:31 UTC
I don't know if it's still the same problem but now it always happens *very* quickly after starting the app. Here's a fresh api trace: https://haagch.frickel.club/files/interface-2.trace.xz
Comment 6 Christoph Haag 2017-08-19 00:50:14 UTC
Just tried it again and it does not hang the GPU now.

They changed the starting environment to another one and had many commits since I last tried, so could be the driver or the app who fixed it.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.