Bug 103130

Summary: Mesa from git causes games to crash mid-play
Product: Mesa Reporter: Adam <padreadamo>
Component: Drivers/Gallium/radeonsiAssignee: Default DRI bug account <dri-devel>
Status: RESOLVED WORKSFORME QA Contact: Default DRI bug account <dri-devel>
Severity: normal    
Priority: medium Keywords: regression
Version: git   
Hardware: x86-64 (AMD64)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:

Description Adam 2017-10-07 00:55:45 UTC
This is my first bug post and I'm a newer Linux user so I hope I did this correctly.


mesa: for the -simplifycfg-sink-common option: may only occur zero or one times!
Comment 1 Adam 2017-10-07 01:17:03 UTC
Downgrading to 17.2-2 (from 17.3 git) and LLVM 5. rectifies the issue. Possible regression?
Comment 2 Kai 2017-10-07 09:10:51 UTC
Hi Adam,
reporting bugs is good! But for them to get resolved, you need to post some more information:
- What GPU are you using? (you can find that in the output of lspci and/or glxinfo)
- Which exact version of Mesa (Git commit) are you using? The same goes for LLVM and Kernel. Did you build these yourself or are you using packages from somewhere?
- What game are you talking about?

Since you report a regression (a downgrade helps you), you should note exactly what components you're downgrading (ideally with exact versions/Git commits). For those components you should do a git bisect to find the offending commit, that introduced the crash for you. For example: if you're saying Mesa 17.2.0 is working and Git HEAD is not, then you would need to do a bisection between these two points. You should end up with the offending commit in the end. If you changed more than one component (eg. LLVM and Mesa) in your downgrade you will most likely need to first find the actual component causing this issue.
Also, since you're reporting a crash, a backtrace might be helpful to see, where the crash occurs.
Comment 3 Adam 2017-10-08 00:33:20 UTC
- What GPU are you using? (you can find that in the output of lspci and/or glxinfo)

VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 470/480/570/580] (rev e7)

- Which exact version of Mesa (Git commit) are you using? The same goes for LLVM and Kernel. Did you build these yourself or are you using packages from somewhere?

I did not build these myself. I'm using a Repo on OpenSuse. They are build from Git. LLVM 5, and Kernel 4.13.4-1

- What game are you talking about?

A wine game, Guild Wars 2, using Gallium Direct X9 state tracker.

I will do some googling to figure out how I can go about bisecting and backtracing. Those are two things I'm unfamiliar with. I did change more than one component through the YAST tool, I switched to a stable repo which uses stable Mesa 17.2 and LLVM 5.0. The error reported by wine specifically states "Mesa," as I copied and pasted that from log.

Thank you for the kind words :-)
Comment 5 Adam 2017-10-17 01:19:37 UTC
I upgraded to the offending repo. I attempted again with wine log -all. I get: 

mesa: for the -simplifycfg-sink-common option: may only occur zero or one times!
/bin/sh: line 1:  4553 Segmentation fault      (core dumped) '/usr/bin/wine' 'Gw2-64.exe'2>&1
Comment 6 Jan Vesely 2017-10-17 11:15:31 UTC
(In reply to Adam from comment #5)
> I upgraded to the offending repo. I attempted again with wine log -all. I
> get: 
> 
> mesa: for the -simplifycfg-sink-common option: may only occur zero or one
> times!
> /bin/sh: line 1:  4553 Segmentation fault      (core dumped) '/usr/bin/wine'
> 'Gw2-64.exe'2>&1

This usually indicates error in LLVM initialization. Do you have multiple LLVM versions installed?
Comment 7 Nicolai Hähnle 2017-10-17 20:36:33 UTC
Thanks for the report!

The error message is benign. The only possible way in which it cause issues is if the application itself is also trying to use LLVM.

If you could provide a backtrace with debug symbols, that would be helpful.
Comment 8 Adam 2017-10-26 01:26:31 UTC
LLVM is certainly the issue. I held back that update, and I didn't get the error. Then pushed the update and received it again going from LLVM 5.0 to LLVM 6.0. However, now the issue seems to be resolved completely. Not quite sure considering no update was pushed.

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.