| Priority | medium |
|---|---|
| Bug ID | 83319 |
| Assignee | dri-devel@lists.freedesktop.org |
| Summary | [r600g] GPU lockup in gsraytrace (Mesa-demo-8.2.0) - RV730 |
| Severity | critical |
| Classification | Unclassified |
| OS | Linux (All) |
| Reporter | Dieter@nuetzel-hh.de |
| Hardware | x86 (IA32) |
| Status | NEW |
| Version | git |
| Component | Drivers/Gallium/r600 |
| Product | Mesa |
Created attachment 105518 [details] dmesg-3.16.1-7.g90bc0f1-desktop-gsraytrace.log GL_RENDERER = Gallium 0.4 on AMD RV730 GL_VERSION = 3.0 Mesa 10.4.0-devel (git-5598458) GL_VENDOR = X.Org GShader thing? mesa-demos/glsl> ./vsraytrace ATTENTION: default value of option vblank_mode overridden by environment. GL_RENDERER = Gallium 0.4 on AMD RV730 47.590482 FPS (238 frames in 5.001000 seconds) 62.325210 FPS (312 frames in 5.006000 seconds) 65.407555 FPS (329 frames in 5.030000 seconds) mesa-demos/glsl> ./fsraytrace ATTENTION: default value of option vblank_mode overridden by environment. GL_RENDERER = Gallium 0.4 on AMD RV730 503.000000 FPS (2515 frames in 5.000000 seconds) 626.800000 FPS (3134 frames in 5.000000 seconds) 577.600000 FPS (2888 frames in 5.000000 seconds) mesa-demos/glsl> ./gsraytrace ATTENTION: default value of option vblank_mode overridden by environment. GL_RENDERER = Gallium 0.4 on AMD RV730 ESC = exit demo left mouse + drag = rotate camera 0.044960 FPS (1 frames in 22.242000 seconds) 0.047201 FPS (1 frames in 21.186000 seconds) ESC / CNTRL+C (several/hundred times) => 2 times switching between blank (black) full screen (console) and desktop (KDE 4.13.3) before the system comes back NO, no LLVM this time... (Michel?) Maybe related: bug 76394
Created att=
achment 105519 [details]
Xorg.0.log-3.16.1-7.g90bc0f1-desktop
Yeah, texturing in geometry shaders hangs. The problem might b= e that the GS RING constant buffer is bound to slot 16, which isn't a valid constant buff= er slot (the last one is 15).
(In reply to comment #2= ) > Yeah, texturing in geometry shaders hangs. The p= roblem might be that the GS > RING constant buffer is bound to slot 16, which isn't a valid constant > buffer slot (the last one is 15). Hello Marek! We're back from vacation, so...;-) Read about it (R600_MAX_CONST_BUFFERS) in Glenn Kennard <glenn.kennard@gmail.= com> patch: [Mesa-dev] [PATCH] r600g: Implement GL_ARB_sample_shading Is it that what you mean? Either way, following the shader dumps. BTW I have a broken screen shot for geom-outlining-150.png, too if you need= .
Created attachme=
nt 105548 [details]
gsraytrace-R600_DEBUG-ps-vs.log
Created attachment=
105549 [details]
gsraytrace-MESA_GLSL-dump.log
(In reply to comment #3= ) > (In reply to comment #2) > > Yeah, texturing in geometry shaders hangs. The problem might be t= hat the GS > > RING constant buffer is bound to slot 16, which isn't a valid con= stant > > buffer slot (the last one is 15). >=20 > Hello Marek! >=20 > We're back from vacation, so...;-) >=20 > Read about it (R600_MAX_CONST_BUFFERS) in Glenn Kennard > <glenn.kennard@g= mail.com> patch: > [Mesa-dev] [PATCH] r600g: Implement GL_ARB_sample_shading >=20 > Is it that what you mean? Yes, but it's just a guess. It may or may not have anything to do with this bug.
| What | Removed | Added |
|---|---|---|
| CC | lil_tux@web.de |
Created attachment 118054 [details]
kmsg: gsraytrace lockup on HD6850
Just noticed the problem today on my HD6850. Unfortunately, it doesn't come
back to life all the time. That is either complete lockup (and reboot) or at
least Xorg being hung in process state D and gsraytrace in state Z. Attached
log produced, when hitting the latter one (though sysrq still worked and
terminal after sysrq-r worked well). The lock contains some of the
sysrq-show-xyz triggers, which are of help maybe (the blocked cat is from
catting one of the radeon sysfs entries). Xorg log only shows event overflows.
GPU hang seems to be 100% reproducable with gsraytrace for me.
Software components being:
media-libs/mesa-9999 (git-4de86e1)
sys-devel/llvm-3.6.2
sys-kernel/vanilla-sources-4.2.0
x11-libs/libdrm-2.4.64
options radeon audio=1 tv=0 dpm=1
(In reply to Heiko from comment #9) > Created attachment 118054 [details] > kmsg: gsraytrace lockup on HD6850 >=20 > Just noticed the problem today on my HD6850. Unfortunately, it doesn't= come > back to life all the time. That is either complete lockup (and reboot)= or at > least Xorg being hung in process state D and gsraytrace in state Z. At= tached > log produced, when hitting the latter one (though sysrq still worked a= nd > terminal after sysrq-r worked well). The lock contains some of the > sysrq-show-xyz triggers, which are of help maybe (the blocked cat is f= rom > catting one of the radeon sysfs entries). Xorg log only shows event > overflows. >=20 > GPU hang seems to be 100% reproducable with gsraytrace for me. >=20 > Software components being: > media-libs/mesa-9999 (git-4de86e1) > sys-devel/llvm-3.6.2 > sys-kernel/vanilla-sources-4.2.0 > x11-libs/libdrm-2.4.64 >=20 > options radeon audio=3D1 tv=3D0 dpm=3D1 Hello Heiko, I've started this one year ago for _RV730_ (AGP) and it is still open for i= t. But as you're on HD6850 and I'm on Turks (6670), now I've opened up new one for those, here: bug 91865 I think you should move your logs there. My impression is that they could be related.
(In reply to Dieter N=C3=BCtzel from comment #3) > (In reply to comment #2) > > Yeah, texturing in geometry shaders hangs. The problem might be t= hat the GS > > RING constant buffer is bound to slot 16, which isn't a valid con= stant > > buffer slot (the last one is 15). >=20 > Hello Marek! >=20 > We're back from vacation, so...;-) >=20 > Read about it (R600_MAX_CONST_BUFFERS) in Glenn Kennard > <glenn.kennard@g= mail.com> patch: > [Mesa-dev] [PATCH] r600g: Implement GL_ARB_sample_shading >=20 > Is it that what you mean? > Either way, following the shader dumps. >=20 > BTW I have a broken screen shot for geom-outlining-150.png, too if you= need. Broken rendering for 'geom-outlining-150' and ogl-samples: 'gl-320-primitive-shading' is FIXED with Mesa-11.0.6 (git, too) on RV730 AGP at least.
'gsraytrace' GPU hang with RV730 (AGP) and NI/Turks XT (6670) still there.
(In reply to Heiko from comment #13) > Try disabling sb for the gs case. Seems to b0rke= n things... Thanks, Heiko. I try normally both versions. ;-) Look here, too: Bug 91865 I have an apitrace, there. Both lookup. But with R600_DEBUG=3Dnosb some seconds (after some broken frames) later. With sb it lookup immediately.
| What | Removed | Added |
|---|---|---|
| See Also | https://bugs.freedesktop.org/show_bug.cgi?id=3D93706 |
Update, this one is NOT solved with current Mesa git. For EG+ (Bug 91865) this is _fixed_ by 'acci= dent' in Mesa git since: commit 2239f3eaff5c72c4cb1d4a5be97feb4af3d08d25 Author: Dave Airlie <airlied&#= 64;redhat.com> Date: Mon Nov 30 15:48:22 2015 +1000 r600/shader: emit tessellation factors to GDS at end of TCS. When we are finished the shader, we read back all the tess factors from LDS and write them to special global memory storage using GDS instructions. This also handles adding NOP when GDS or ENDLOOP end the TCS. Signed-off-by: Dave Airlie <airlied@redhat.com> Dave and Marek any hints which could point in the right direction? What is different in this case between R600/R700 and EG+ (NI/Turks in my ca= se) and what should I try next.
I have reported probably a duplicate: https://bugs.freedesktop.org/show_bug.c= gi?id=3D93706 It has also an apitrace trace. If confirmed, i will mark it as a duplicate<= /pre>
| What | Removed | Added |
|---|---|---|
| Resolution | --- | MOVED |
| Status | NEW | RESOLVED |
-- 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/525.