Bug 111014

Created attachment 144658 [details]
output of sysfs error file

Created attachment 144659 [details]
output sysfs Kernel 5.2rc5

Created attachment 144660 [details]
output sysfs Kernel 5.2rc4

Created attachment 144661 [details]
output sysfs Kernel 5.2rc3

Created attachment 144662 [details]
output sysfs Kernel 5.1rc1

Ok, we'll do before and after! Hopefully everyone will be happy!

Created attachment 144663 [details]
dmesg error Kernel 5.2rc6

Created attachment 144664 [details]
dmesg error Kernel 5.2rc5

Please try:

diff --git a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
index 81f9b0422e6a..f11ba6da4d1d 100644
--- a/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/gt/intel_ringbuffer.c
@@ -1811,6 +1811,11 @@ static int ring_request_alloc(struct i915_request *request)
        if (ret)
                return ret;
 
+       /* Once again for Ivybridge after updating 3D state. */
+       ret = request->engine->emit_flush(request, EMIT_INVALIDATE);
+       if (ret)
+               return ret;
+
        request->reserved_space -= LEGACY_REQUEST_SIZE;
        return 0;
 }

Created attachment 144665 [details]
dmesg error Kernel 5.2rc4

Created attachment 144666 [details]
dmesg error Kernel 5.2rc3

Created attachment 144667 [details]
dmesg error Kernel 5.1rc1

Created attachment 144668 [details]
dmesg error Kernel 5.1.12

Created attachment 144669 [details]
output sysfs Kernel 5.1.12

Created attachment 144670 [details]
dmesg error Kernel f2253bd9859b - bad commit in bisect log

Created attachment 144671 [details]
output sysfs Kernel f2253bd9859b - bad commit in bisect log

Waitasec, in upstream, we invalidate before the switch. Could you please check with 5.2 to see if it is already fixed?

See commit 928f8f42310f244501a7c70daac82c196112c190
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Apr 19 12:17:47 2019 +0100

    drm/i915/ringbuffer: EMIT_INVALIDATE *before* switch context
    
    Despite what I think the prm recommends, commit f2253bd9859b
    ("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context") turned out
    to be a huge mistake when enabling Ironlake contexts as the GPU would
    hang on either a MI_FLUSH or PIPE_CONTROL immediately following the
    MI_SET_CONTEXT of an active mesa context (more vanilla contexts, e.g.
    simple rendercopies with igt, do not suffer).

(In reply to Chris Wilson from comment #17)
> Waitasec, in upstream, we invalidate before the switch. Could you please
> check with 5.2 to see if it is already fixed?

it happens in 5.2rc1 to the latest 5.2-rc6 too - see the attached files of the dmesg output and the sysfs file

i attached 5.1.12 (so after release), 5.1-rc1 of 5.1 series
AND
various 5.2-rcX messages/error sysfs files
and a log+sysfs from the first bad commit I found during bisect

PS: drivers/gpu/drm/i915/gt/intel_ringbuffer.c does not exist in Linux 5.2-rc6 so I cannot test that - and it does not seem to apply if i try to patch the i915/intel_ringbuffer.c that is there

You need our version of 5.2 ;) https://cgit.freedesktop.org/drm-tip

(In reply to Chris Wilson from comment #20)
> You need our version of 5.2 ;) https://cgit.freedesktop.org/drm-tip

Thank you. I modified my PKGBUILD and created a new kernel pkg with that patch file applied (because the source tree has the patched version).

reports now as 5.2.0-rc6-bisect-g44b3a556c682 

Preliminary: the patch seems to fix the issue / it doesnt occur w. that patched kernel variant

Detail: 4 hours uptime with some video playback loop in mpv and active chromium w. youtube is fine - the error was manifesting earlier in all other cases in my logs but I will run it a little while longer to be sure; I didnt test the kernel without that patch

Hello, could you please clarify how many RAM do you have on your PC? Asking because looks like we also could reproduce it, but only when removed 4 GB memory (4 left). On monday will check suggested by Chris kernel version.

(In reply to Denis from comment #22)
> Hello, could you please clarify how many RAM do you have on your PC? 

8GB RAM

running currently with kernel cmdline (also see dmesg): rw verbose sysrq_always_enabled audit=0 intel_iommu=on,igfx_off

I think active IOMMU (for VT-d) isnt normally enabled.
Also using 2 monitors (see dmesg) with

xrandr --listmonitors
Monitors: 2
 0: +*HDMI-2 1680/474x1050/296+1920+0  HDMI-2
 1: +HDMI-1 1920/521x1080/293+0+0  HDMI-1


> Preliminary: the patch seems to fix the issue

Final: the patch seems to fix the issue / it doesnt occur w. that patched kernel variant

After now little more than 13 hours running video in a loop there is no error related to this bug.

On its way back to v5.1:

commit c84c9029d782a3a0d2a7f0522ecb907314d43e2c
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Apr 19 12:17:47 2019 +0100

    drm/i915/ringbuffer: EMIT_INVALIDATE *before* switch context
    
    Despite what I think the prm recommends, commit f2253bd9859b
    ("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context") turned out
    to be a huge mistake when enabling Ironlake contexts as the GPU would
    hang on either a MI_FLUSH or PIPE_CONTROL immediately following the
    MI_SET_CONTEXT of an active mesa context (more vanilla contexts, e.g.
    simple rendercopies with igt, do not suffer).
    
    Ville found the following clue,
    
      "[DevCTG+]: For the invalidate operation of the pipe control, the
       following pointers are affected. The
       invalidate operation affects the restore of these packets. If the pipe
       control invalidate operation is completed
       before the context save, the indirect pointers will not be restored from
       memory.
       1. Pipeline State Pointer
       2. Media State Pointer
       3. Constant Buffer Packet"
    
    which suggests by us emitting the INVALIDATE prior to the MI_SET_CONTEXT,
    we prevent the context-restore from chasing the dangling pointers within
    the image, and explains why this likely prevents the GPU hang.
    
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
    Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
    Link: https://patchwork.freedesktop.org/patch/msgid/20190419111749.3910-1-chris@chris-wilson.co.uk
    (cherry picked from commit 928f8f42310f244501a7c70daac82c196112c190 in drm-intel-next)
    Cc: stable@vger.kernel.org
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111014
    Fixes: f2253bd9859b ("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context")
    Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

*** Bug 110912 has been marked as a duplicate of this bug. ***

*** Bug 110969 has been marked as a duplicate of this bug. ***

*** Bug 110985 has been marked as a duplicate of this bug. ***

*** Bug 110867 has been marked as a duplicate of this bug. ***

*** Bug 110860 has been marked as a duplicate of this bug. ***

*** Bug 110858 has been marked as a duplicate of this bug. ***

*** Bug 110834 has been marked as a duplicate of this bug. ***

*** Bug 110816 has been marked as a duplicate of this bug. ***

*** Bug 110812 has been marked as a duplicate of this bug. ***

*** Bug 110800 has been marked as a duplicate of this bug. ***

*** Bug 110652 has been marked as a duplicate of this bug. ***

Hi Chris
I've checked the issues (which you've marked as duplicates) on the 5.2.0 version of drm-tip - the issue isn't reproduced.
Also, I've checked the issue on the previous commit of the drm-tip, the issue is also not relevant .
Therefore, the issue was fixed with one of the previous commits or its some kind of complex fix.

(In reply to Chris Wilson from comment #18)
> See commit 928f8f42310f244501a7c70daac82c196112c190
> Author: Chris Wilson <chris@chris-wilson.co.uk>
> Date:   Fri Apr 19 12:17:47 2019 +0100
> 
>     drm/i915/ringbuffer: EMIT_INVALIDATE *before* switch context
>     
>     Despite what I think the prm recommends, commit f2253bd9859b
>     ("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context") turned out
>     to be a huge mistake when enabling Ironlake contexts as the GPU would
>     hang on either a MI_FLUSH or PIPE_CONTROL immediately following the
>     MI_SET_CONTEXT of an active mesa context (more vanilla contexts, e.g.
>     simple rendercopies with igt, do not suffer).

This fixes super-annoying hangs on Ivy Bridge with v5.1.15. Thanks for the pointer!