Bug 106194

This happened to me too: the same BUG() with the same stacktrace, on 4.17-rc2.  It was when I was away, so maybe it happened when the screen turned off due to inactivity.  Graphics card is an Radeon RX 550, one monitor connected via HDMI.  Didn't happen with 4.16.  I did not test 4.17-rc1.

Created attachment 139197 [details]
dmesg output linux 4.17 rc2

Comment on attachment 139197 [details]
dmesg output linux 4.17 rc2

This is from one boot to the next boot. I booted the system at 18:22 and I was using it for upto 20:00. I went for a cup of tea and system froze at 20:08.

Then I've booted the system again.

I am using a hand compiled linux kernel 4.17-rc2

I know that I left the system at 20:00

Apr 28 20:08:28 levis-desktop kernel: kernel BUG at drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c:4700!
Apr 28 20:08:28 levis-desktop kernel: RIP: 0010:dm_update_crtcs_state+0x347/0x3c0 [amdgpu]
Apr 28 20:08:28 levis-desktop kernel:  amdgpu_dm_atomic_check+0x191/0x3e0 [amdgpu]
Apr 28 20:08:28 levis-desktop kernel:  amdgpu_drm_ioctl+0x49/0x80 [amdgpu]
Apr 28 20:08:28 levis-desktop kernel: RIP: dm_update_crtcs_state+0x347/0x3c0 [amdgpu] RSP: ffffa87b814bbb18


System:    Host: levis-desktop Kernel: 4.17.0-rc2-next-20180426-ARCH x86_64 bits: 64 Desktop: KDE Plasma 5.12.4 
           Distro: Arch Linux 
Machine:   Type: Desktop Mobo: Micro-Star model: A320M PRO-VD PLUS (MS-7B38) v: 1.0 serial: N/A 
           UEFI: American Megatrends v: 1.80 date: 03/15/2018 
CPU:       Topology: Quad Core model: AMD Ryzen 5 2400G with Radeon Vega Graphics bits: 64 type: MT MCP 
           L2 cache: 2048 KiB 
           Speed: 1368 MHz min/max: 1600/3600 MHz Core speeds (MHz): 1: 1611 2: 1381 3: 1373 4: 1470 5: 1834 
           6: 2727 7: 1375 8: 1376 
Graphics:  Card-1: AMD Raven Bridge [Radeon Vega Series / Radeon Vega Mobile Series] driver: amdgpu v: kernel 
           Display: x11 server: X.Org 1.19.6 driver: modesetting unloaded: ati,fbdev,vesa 
           resolution: 1600x900~60Hz 
           OpenGL: renderer: AMD RAVEN (DRM 3.25.0 / 4.17.0-rc2-next-20180426-ARCH LLVM 6.0.0) 
           v: 4.5 Mesa 18.0.1

I might be having the same issue, however my stack looks a bit different, perhaps due to different kernel commits and the way it is triggered. Can you guys trigger this hang by locking the window manager session/screen and waiting for the display(s) to go to sleep? That's how I can reliably trigger my similar hang.

april 28 20:16:49 beist.localdomain kernel: kernel BUG at drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c:4708!                                                                                        
april 28 20:16:49 beist.localdomain kernel: invalid opcode: 0000 [#1] SMP NOPTI 
april 28 20:16:49 beist.localdomain kernel: RIP: 0010:dm_update_crtcs_state+0x397/0x410 [amdgpu] 
...
april 28 20:16:49 beist.localdomain kernel: Call Trace:                                                                                                                                                            
april 28 20:16:49 beist.localdomain kernel:  amdgpu_dm_atomic_check+0x182/0x3b0 [amdgpu]                                                                                                                           
april 28 20:16:49 beist.localdomain kernel:  drm_atomic_check_only+0x33a/0x4f0 [drm]                                                                                                                               
april 28 20:16:49 beist.localdomain kernel:  drm_atomic_commit+0x13/0x50 [drm]                                                                                                                                     
april 28 20:16:49 beist.localdomain kernel:  drm_atomic_connector_commit_dpms+0xe5/0xf0 [drm]                                                                                                                      
april 28 20:16:49 beist.localdomain kernel:  drm_mode_obj_set_property_ioctl+0x174/0x290 [drm]                                                                                                                     
april 28 20:16:49 beist.localdomain kernel:  ? drm_mode_connector_set_obj_prop+0x70/0x70 [drm]                                                                                                                     
april 28 20:16:49 beist.localdomain kernel:  drm_mode_connector_property_set_ioctl+0x3e/0x60 [drm]                                                                                                                 
april 28 20:16:49 beist.localdomain kernel:  drm_ioctl_kernel+0x5b/0xb0 [drm]                                                                                                                                      
april 28 20:16:49 beist.localdomain kernel:  drm_ioctl+0x2c3/0x360 [drm]                                                                                                                                           
april 28 20:16:49 beist.localdomain kernel:  ? drm_mode_connector_set_obj_prop+0x70/0x70 [drm]                                                                                                                     
april 28 20:16:49 beist.localdomain kernel:  amdgpu_drm_ioctl+0x49/0x80 [amdgpu]                                                                                                                                   
april 28 20:16:49 beist.localdomain kernel:  do_vfs_ioctl+0xa4/0x620                                                                                                                                               
april 28 20:16:49 beist.localdomain kernel:  ksys_ioctl+0x70/0x80                                                                                                                                                  
april 28 20:16:49 beist.localdomain kernel:  __x64_sys_ioctl+0x16/0x20                                                                                                                                             
april 28 20:16:49 beist.localdomain kernel:  do_syscall_64+0x5b/0x160                                                                                                                                              
april 28 20:16:49 beist.localdomain kernel:  entry_SYSCALL_64_after_hwframe+0x44/0xa9                                                                                                                              
april 28 20:16:49 beist.localdomain kernel: RIP: 0033:0x7f7d092cc0f7

Created attachment 139205 [details] [review]
Partially revert commit introducing the BUG() where the "hang" happens

This is the patch I'm currently testing to avoid the hang in my case. It probably does not solve the issue, but it seem to avoid the hang for me. It's just that my original problem is that after the displays go to sleep the stay blank until I reboot, even with the patch.

For me also, the issues occurs when screen's display go black for inactivity and a hang on wake.

Is this happening in 4.17rc3?

Seems to be resolved with 4.17rc3 so far!

This just happened to me again, now on v4.17-rc4.  It's the 'BUG_ON(dm_new_crtc_state->stream == NULL);' at amdgpu_dm.c:4708, apparently happened when the screen turned off due to inactivity.  Graphics card is an Radeon RX 550, one monitor connected via HDMI.

I'm having trouble reproducing the BUG_ON in Ubuntu 18.04.
For those who can, the below info will be helpful:

- Distribution & version
- Window manager used
- Xorg log
- journalctl log, if on Wayland

Xorg log is at /var/log/Xorg.x.log
Gzip journalctl via `$ journalctl | gzip - > journalctl.gz

Created attachment 139602 [details] [review]
Fix v1

It seems I've reproduced it by using the default modesetting DDX driver.
Please give the attached patch a shot.

Confirming i had the same issue on suspend. Patch "Fix v1" Resolves the issue for me. Attaching "before" and "after" dmesg outputs.

Created attachment 139634 [details]
"Before fix" dmesg output

Created attachment 139635 [details]
"After Fix" dmesg output

Note a "new" error near the end. No loss in functionality with the error, screen behaves normally. Cannot confirm if new trace is in response to the patch or not.  The 84 second range is right at the time i blanked and un-blanked the screen. I was using ``xset dpms force off".

Just confirmed that the error in the "after" dmesg output occurs every time the screen is blanked. Will post a new bug once the patch listed here is mainlined.

So which rc version will contain the fix?

(In reply to Shmerl from comment #17)
> So which rc version will contain the fix?

It's already in Dave's drm-fixes: https://cgit.freedesktop.org/~airlied/linux/log/?h=drm-fixes

So it should make it to the 4.17 release.

(In reply to Leo Li from comment #18)
> (In reply to Shmerl from comment #17)
> > So which rc version will contain the fix?
> 
> It's already in Dave's drm-fixes:
> https://cgit.freedesktop.org/~airlied/linux/log/?h=drm-fixes
> 
> So it should make it to the 4.17 release.

I can reproduce this (more precisely bug 104611, a likely duplicate) on 4.17.5.