Bug 109539

Summary: System freezing
Product: DRI Reporter: jon <jon780>
Component: DRM/AMDgpuAssignee: Default DRI bug account <dri-devel>
Status: NEW --- QA Contact:
Severity: normal    
Priority: medium CC: harry.wentland, nicholas.kazlauskas
Version: XOrg git   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
journalctl
none
dmesg output
none
Xorg.0.log none

Description jon 2019-02-03 20:36:56 UTC
Created attachment 143282 [details]
journalctl

Fedora 29
kernel 4.20.5 (compiled to try something newer than in the fedora repos)
Ryzen 1700X
MSI B350 Tomawhawk (Arctic)
Three Dell U2415 monitors (one dvi, one hdmi, one displayport).

Freezing when waking from sleep, usually withing just a few seconds.  Example journactl attached.  When it freezes, the mouse can still move, but no other input.  Cannot click, cannot switch to virtual terminals, cannot send REISUB, cannot toggle off numlock (light frozen).  Only thing that fixes it is a reboot of the computer.

Setting amdgpu.dc=0 appears to resolve the issue!  Went from freezing when waking from sleep several times a day to >24 hours of uptime so far.

Also, when DC is enabled, displayport monitor will not wake from sleep.  The only want I can get it to wake is xrandr --off and then --auto on the DisplayPort.  I also want to mention that the displayport name changes every time it sleeps and wakes (DisplayPort-1, DisplayPort-2, DisplayPort-3) even though nothing is really being moved to a different port.
Comment 1 Nicholas Kazlauskas 2019-02-04 15:35:30 UTC
The freezing doesn't seem to be necessarily caused by by something in amdgpu from a glance at your log.

There are some DC warnings/errors that describe what you're seeing with your monitor not waking from sleep however.

Please post a full dmesg log from system boot and an xorg log if you're using X. It may also help to know what window manager you're using when you see this issue occur.
Comment 2 Alex Deucher 2019-02-04 19:02:51 UTC
Does adding idle=nomwait to the kernel command line in grub help?
Comment 3 jon 2019-02-04 23:49:37 UTC
>The freezing doesn't seem to be necessarily caused by by something in amdgpu >from a glance at your log.

I thought the same thing, but after disabling direct code I went from the system freezing up to 5 or 6 times a day, depending on use, every day to having an uptime over 2 days and 4 hours now.  So there's no question that disabling direct code solved the problem.  It is night and day after that change.

>There are some DC warnings/errors that describe what you're seeing with your monitor not waking from sleep however.
>
>Please post a full dmesg log from system boot and an xorg log if you're using X. >It may also help to know what window manager you're using when you see this >issue occur.

I am using X and my window manager is dwm.  I will attach an Xorg log as well as dmesg.  Keep in mind these will be the most recent logs with direct code disabled, which may not matter, just wanted to be clear.  

>Does adding idle=nomwait to the kernel command line in grub help?

No, according to my notes I added idle=nomwait on 1/31/19 and experienced many freezes per day since then.  I also used the following kernel boot parameters, which had no affect:

rcu_nocbs=0-15
processor.max_cstate=5

All three of these, along with disabling direct code, are still in my kernel boot parameters.  None of those had any affect, frequent crashes still.  After disabling direct code and restarting I have not had one freeze yet.  Stability went from a couple of hours (usually long enough for me to be away from it long enough to sleep) to an uptime of >2 days.
Comment 4 jon 2019-02-04 23:52:24 UTC
Created attachment 143290 [details]
dmesg output

dmesg output (direct code disabled)
Comment 5 jon 2019-02-04 23:53:12 UTC
Created attachment 143291 [details]
Xorg.0.log

xorg log from ~/.local/
Comment 6 Nicholas Kazlauskas 2019-02-05 17:07:40 UTC
Do you still see the issue occur when amdgpu.dc=1 if you disable DP1.2 support in your monitor's OSD?
Comment 7 jon 2019-02-10 14:40:48 UTC
>Do you still see the issue occur when amdgpu.dc=1 if you disable DP1.2 
>support in your monitor's OSD?

I will let you know.  After 7+ days of uptime without a single crash with amdgpu.dc=0 I updated from kernel 4.20.5 to 4.20.7, set amdgpu.dc=1 and disabled displayport 1.2 on the monitor using displayport.  I will report back.
Comment 8 jon 2019-02-12 01:45:41 UTC
I think disabling displayport and leaving direct code enabled is stable as well.  Up to 1 day and 11 hours of uptime.
Comment 9 jon 2019-02-12 01:46:59 UTC
Sorry, to clarify, I disabled displayport 1.2.

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.