Bug 90284

Summary: GPU lockup with DOTA2
Product: Mesa Reporter: Trippler <niklas>
Component: Drivers/Gallium/radeonsiAssignee: Default DRI bug account <dri-devel>
Status: RESOLVED INVALID QA Contact: Default DRI bug account <dri-devel>
Severity: normal    
Priority: medium CC: daniel, niklas
Version: 10.5   
Hardware: x86-64 (AMD64)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments: journalctl -r -x --boot=-4
journalctl -r -x --boot=-2. This is the log when booted with radeon.dpm=0
Output of the program being run with R600_DEBUG=ps,vs,gs
journalctl -r -x --boot=-1 when recreating the crash with R600_DEBUG=ps,vs,gs

Description Trippler 2015-05-02 17:23:48 UTC
Created attachment 115515 [details]
journalctl -r -x --boot=-4

Graphics card: AMD R9 290
Kernel: Linux archfractal 4.0.1-1-ARCH #1 SMP PREEMPT
Monitors connected: 2
DE: Xfce

Relevant packages:
local/xf86-video-ati 1:7.5.0-2 (xorg-drivers xorg)
local/lib32-mesa 10.5.4-1
local/lib32-mesa-libgl 10.5.4-1
local/lib32-mesa-vdpau 10.5.4-1
local/linux 4.0.1-1 (base)
local/xorg-server 1.17.1-5 (xorg)

This bug, which is triggered by dota2 (Valve game), locks up both displays for a few seconds before they blink black and freeze in the last frame displayed, requireing a system reboot. The bug seems to happen with others (same symptoms / log messages) using other applications (chrome html5 video was mentioned), but dota2 seems like one of the few recreatable ways to trigger the bug.

Steps to recreate:
Have steam and dota 2 installed (might have to remove steams own bundled libstdc++ libraries as they conflict with mesa).
Launch dota 2 with any video settings
Start a new lobby with "cheats enabled" and "all pick" game mode (these don't matter, but make it faster to recreate)
Pick the hero "Chaos Knight" and enter the game
Type in chat "-lvlup 24" to gain max level and skill up everything
Click R to use ulti (which spawns copies of himself)
Hold-drag mouse over all the illusions and the hero.
When the mouse button is released the above bug triggers, crashing the displaydriver and taking the computer down with it.

In the log attached (journal_log) the bug is triggered at 17:43:50, the first line there is: "kernel: radeon 0000:01:00.0: GPU fault detected: 147 0x000cc801"
I noticed a few lines later the following lines appeared:

May 02 17:44:01 archfractal kernel: [drm:radeon_pm_resume [radeon]] *ERROR* radeon: dpm resume failed
May 02 17:44:01 archfractal kernel: [drm:ci_dpm_enable [radeon]] *ERROR* ci_start_dpm failed

I guessed the bug had something to do with dpm, and disabled it (kernel parameter radeon.dpm=0). When I had done this the bug could still be recreated, but the log showed completely different error messages. This log is also attached as "journal_log_radeon_dpm_is_0"

I had to gzip the first log to fit within the 3M file size limit
Comment 1 Trippler 2015-05-02 17:24:50 UTC
Created attachment 115516 [details]
journalctl -r -x --boot=-2. This is the log when booted with radeon.dpm=0
Comment 2 Tom Stellard 2015-05-06 14:16:43 UTC
Can you run the program with the environment variable R600_DEBUG=ps,vs,gs and post the output.
Comment 3 Trippler 2015-05-06 14:43:30 UTC
Created attachment 115594 [details]
Output of the program being run with R600_DEBUG=ps,vs,gs

export R600_DEBUG=ps,vs,gs
steam > output_steam

After this I ran game from steam and recreated crash condition. Also attached (next attachment) is systemd log from this event as well (journal_log_with_debug_ps_vs_gs.gz)
Comment 4 Trippler 2015-05-06 14:44:14 UTC
Created attachment 115595 [details]
journalctl -r -x --boot=-1 when recreating the crash with R600_DEBUG=ps,vs,gs
Comment 5 Trippler 2015-05-06 14:45:16 UTC
(In reply to Trippler from comment #4)
> Created attachment 115595 [details]
> journalctl -r -x --boot=-1 when recreating the crash with R600_DEBUG=ps,vs,gs

Done. I added two logs - "output_steam.gz" which is the output of the program, and "journal_log_with_debug_ps_vs_gs.gz" which is the output of "journalctl --boot=-1 -r -x"
Comment 6 pablow.1422 2015-05-09 10:17:11 UTC
Same here. 

Graphics card: AMD Radeon HD7750
Kernel: 3.19.0-16-generic #16-Ubuntu SMP
DE: KDE

Mesa 10.6.0-devel (git-28090b3 2015-05-05 vivid-oibaf-ppa)

Any info needed, I'm happy to provide it.
Comment 7 Marek Olšák 2015-10-17 17:27:32 UTC
Can you still reproduce this with current Mesa git?
Comment 8 Oded Gabbay 2016-05-24 19:16:49 UTC
Is this still relevant ?
We got reports it works with Ubuntu 16.04
Comment 9 Trippler 2016-05-25 14:30:53 UTC
(In reply to Oded Gabbay from comment #8)
> Is this still relevant ?
> We got reports it works with Ubuntu 16.04

I can confirm that I can no longer recreate the bug. They have switched the game engine the game uses since this bug was reported, so I do not know if the cause is fixed or it is no longer triggered.

I do think it's safe to mark this as no longer relevant though.
Comment 10 Timothy Arceri 2018-04-03 03:31:37 UTC
Closing as invalid since both Mesa and Dota2 have changed since the report and the problem can no longer be reproduced.

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.