Bug 60347

Summary:	Htile support causes GPU lockup CP stall in Amnesia game
Product:	Mesa	Reporter:	Anonymous Helper <anonymous>
Component:	Drivers/Gallium/r600	Assignee:	Default DRI bug account <dri-devel>
Status:	RESOLVED DUPLICATE	QA Contact:
Severity:	normal
Priority:	medium
Version:	git
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:

Description Anonymous Helper 2013-02-06 02:26:54 UTC

There is/are a reproducible GPU lockup in Amnesia: Dark Descent video game on Linux x86-64. Often it occurs when looking at certain doors in the levels. I find that using VT-switch helps to recover the GPU in a reasonable timeframe to a somewhat usable state. I do experience some jerkiness though after the stall recovery.

Radeon 6750M

3.8.0-rc5
Everything else taken from xorg-edgers ppa

I'm happy to investigate this issue further.

[ 9902.661056] radeon 0000:01:00.0: GPU lockup CP stall for more than 10000msec
[ 9902.661059] radeon 0000:01:00.0: GPU lockup (waiting for 0x0000000000049060)
[ 9902.661062] radeon 0000:01:00.0: failed to get a new IB (-35)
[ 9902.661063] [drm:radeon_cs_ib_chunk] *ERROR* Failed to get ib !
[ 9902.662124] radeon 0000:01:00.0: Saved 10007 dwords of commands on ring 0.
[ 9902.662127] radeon 0000:01:00.0: GPU softreset: 0x00000003
[ 9902.673782] radeon 0000:01:00.0:   GRBM_STATUS               = 0xA7732CA4
[ 9902.673784] radeon 0000:01:00.0:   GRBM_STATUS_SE0           = 0x7C000005
[ 9902.673786] radeon 0000:01:00.0:   GRBM_STATUS_SE1           = 0x00000007
[ 9902.673787] radeon 0000:01:00.0:   SRBM_STATUS               = 0x200000C0
[ 9902.673789] radeon 0000:01:00.0:   R_008674_CP_STALLED_STAT1 = 0x04000000
[ 9902.673791] radeon 0000:01:00.0:   R_008678_CP_STALLED_STAT2 = 0x000C0000
[ 9902.673793] radeon 0000:01:00.0:   R_00867C_CP_BUSY_STAT     = 0x00048402
[ 9902.673794] radeon 0000:01:00.0:   R_008680_CP_STAT          = 0x80860243
[ 9902.673796] radeon 0000:01:00.0:   GRBM_SOFT_RESET=0x00007F6B
[ 9902.673848] radeon 0000:01:00.0:   GRBM_STATUS               = 0x00003828
[ 9902.673850] radeon 0000:01:00.0:   GRBM_STATUS_SE0           = 0x00000007
[ 9902.673851] radeon 0000:01:00.0:   GRBM_STATUS_SE1           = 0x00000007
[ 9902.673853] radeon 0000:01:00.0:   SRBM_STATUS               = 0x200000C0
[ 9902.673855] radeon 0000:01:00.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[ 9902.673857] radeon 0000:01:00.0:   R_008678_CP_STALLED_STAT2 = 0x00000000
[ 9902.673858] radeon 0000:01:00.0:   R_00867C_CP_BUSY_STAT     = 0x00000000
[ 9902.673860] radeon 0000:01:00.0:   R_008680_CP_STAT          = 0x00000000
[ 9902.691442] radeon 0000:01:00.0: GPU reset succeeded, trying to resume
[ 9902.709761] [drm] probing gen 2 caps for device 8086:101 = 2/0
[ 9902.709765] [drm] PCIE gen 2 link speeds already enabled
[ 9902.711799] [drm] PCIE GART of 512M enabled (table at 0x0000000000040000).
[ 9902.711901] radeon 0000:01:00.0: WB enabled
[ 9902.711904] radeon 0000:01:00.0: fence driver on ring 0 use gpu addr 0x0000000040000c00 and cpu addr 0xffff880451db8c00
[ 9902.711905] radeon 0000:01:00.0: fence driver on ring 3 use gpu addr 0x0000000040000c0c and cpu addr 0xffff880451db8c0c
[ 9902.728078] [drm] ring test on 0 succeeded in 3 usecs
[ 9902.728142] [drm] ring test on 3 succeeded in 1 usecs
[ 9902.777649] [drm] ib test on ring 0 succeeded in 0 usecs
[ 9902.777738] [drm] ib test on ring 3 succeeded in 1 usecs
[ 9914.739076] [drm] Forcing lvds to dual link mode on Apple MacBook Pro (Core i5/i7 Series)

Comment 1 Michel Dänzer 2013-02-06 09:54:13 UTC

Please attach /var/log/Xorg.0.log and the output of glxinfo and dmesg.

> 3.8.0-rc5
> Everything else taken from xorg-edgers ppa

Does the problem occur with a 3.7 kernel and/or without the PPA?

Comment 2 Anonymous Helper 2013-02-07 12:11:00 UTC

(In reply to comment #1)
> Please attach /var/log/Xorg.0.log and the output of glxinfo and dmesg.
> 
> > 3.8.0-rc5
> > Everything else taken from xorg-edgers ppa
> 
> Does the problem occur with a 3.7 kernel and/or without the PPA?

I've narrowed it down to one of these two kernel patches:

4ac0533abaec2b83a7f2c675010eedd55664bc26
cf4ccd016bae1a03bb38170eb54b5db4b04e0545

Comment 3 Alex Deucher 2013-02-07 13:03:26 UTC

(In reply to comment #2)
> 
> I've narrowed it down to one of these two kernel patches:
> 
> 4ac0533abaec2b83a7f2c675010eedd55664bc26
> cf4ccd016bae1a03bb38170eb54b5db4b04e0545

It's probably a mesa bug then.  Since those two patches just enable mesa to use additional features (htile and DMA).  Probably one of these mesa commits:
24b1206ab2dcd506aaac3ef656aebc8bc20cd27a
6532eb17baff6e61b427f29e076883f8941ae664
325422c49449acdd8df1eb2ca8ed81f7696c38cc

Does setting the env var R600_HYPERZ=0 or resetting your git tree back to cf5632094ba0c19d570ea47025cf6da75ef8457a fix the issue?

Comment 4 Anonymous Helper 2013-02-07 14:45:16 UTC

(In reply to comment #3)
> It's probably a mesa bug then.  Since those two patches just enable mesa to
> use additional features (htile and DMA).  Probably one of these mesa commits:
> 24b1206ab2dcd506aaac3ef656aebc8bc20cd27a
> 6532eb17baff6e61b427f29e076883f8941ae664
> 325422c49449acdd8df1eb2ca8ed81f7696c38cc
> 
> Does setting the env var R600_HYPERZ=0 or resetting your git tree back to
> cf5632094ba0c19d570ea47025cf6da75ef8457a fix the issue?

I couldn't bisect mesa due to compile issues, but disabling HyperZ does fix the issue.

Comment 5 Anonymous Helper 2013-02-08 01:06:12 UTC

This bug looks like to be a duplicate of 59592.

Comment 6 Alex Deucher 2013-02-08 02:10:40 UTC


*** This bug has been marked as a duplicate of bug 59592 ***

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.