Bug 103634

Summary:	nv_conditional_render tests hang the gpu (HD 5750)
Product:	Mesa	Reporter:	Roland Scheidegger <sroland>
Component:	Drivers/Gallium/r600	Assignee:	Default DRI bug account <dri-devel>
Status:	RESOLVED MOVED	QA Contact:	Default DRI bug account <dri-devel>
Severity:	major
Priority:	medium
Version:	git
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:
Attachments:	trace file captured with R600_TRACE

Description Roland Scheidegger 2017-11-09 02:53:38 UTC

Nearly all of piglit's nv_conditional_render tests (I think 10 out of 12, I'd rather not retry them all since it sometimes brings down the box after the gpu reset) immediately hang the gpu - nv_conditional_render-begin-xxx are the two exceptions (which pass).
GPU is Juniper, HD 5750.
I'm not sure if this is even a regression, as I never ran piglit on this box before, and at least with mesa 17.1.8 it hangs all the same as current git master. If the kernel version matters, I tried some 4.9 and some 4.13 kernel.

[  749.782315] radeon 0000:01:00.0: ring 0 stalled for more than 10224msec
[  749.782319] radeon 0000:01:00.0: GPU lockup (current fence id 0x0000000000004081 last fence id 0x0000000000004
...

[  771.466593] radeon 0000:01:00.0: Saved 882 dwords of commands on ring 0.
[  771.906456] radeon 0000:01:00.0: GPU softreset: 0x0000001D
[  771.906458] radeon 0000:01:00.0:   GRBM_STATUS               = 0xA0001828
[  771.906459] radeon 0000:01:00.0:   GRBM_STATUS_SE0           = 0x00000003
[  771.906460] radeon 0000:01:00.0:   GRBM_STATUS_SE1           = 0x00000007
[  771.906461] radeon 0000:01:00.0:   SRBM_STATUS               = 0x20000AC0
[  771.906462] radeon 0000:01:00.0:   SRBM_STATUS2              = 0x00000000
[  771.906463] radeon 0000:01:00.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  771.906464] radeon 0000:01:00.0:   R_008678_CP_STALLED_STAT2 = 0x00000000
[  771.906465] radeon 0000:01:00.0:   R_00867C_CP_BUSY_STAT     = 0x00000006
[  771.906466] radeon 0000:01:00.0:   R_008680_CP_STAT          = 0x80008E47
[  771.906468] radeon 0000:01:00.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  771.914331] radeon 0000:01:00.0: GRBM_SOFT_RESET=0x00007F6B
[  771.914382] radeon 0000:01:00.0: SRBM_SOFT_RESET=0x00100100
[  771.915528] radeon 0000:01:00.0:   GRBM_STATUS               = 0x00003828
[  771.915529] radeon 0000:01:00.0:   GRBM_STATUS_SE0           = 0x00000007
[  771.915530] radeon 0000:01:00.0:   GRBM_STATUS_SE1           = 0x00000007
[  771.915531] radeon 0000:01:00.0:   SRBM_STATUS               = 0x200000C0
[  771.915532] radeon 0000:01:00.0:   SRBM_STATUS2              = 0x00000000
[  771.915533] radeon 0000:01:00.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  771.915534] radeon 0000:01:00.0:   R_008678_CP_STALLED_STAT2 = 0x00000000
[  771.915535] radeon 0000:01:00.0:   R_00867C_CP_BUSY_STAT     = 0x00000000
[  771.915536] radeon 0000:01:00.0:   R_008680_CP_STAT          = 0x00000000
[  771.915538] radeon 0000:01:00.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  771.915554] radeon 0000:01:00.0: GPU reset succeeded, trying to resume

Comment 1 Roland Scheidegger 2017-11-09 04:42:53 UTC

Created attachment 135335 [details]
trace file captured with R600_TRACE

Here's the output of R600_TRACE when running ./nv_conditional_render-vertex_array. (It doesn't tell me much but there is indeed a SET_PREDICATION and a predicated draw between the last trace id reached and the first not reached one.)

Comment 2 GitLab Migration User 2019-09-18 19:24:32 UTC

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/616.

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.