How to reproduce the bug: 1. Get a SKL machine with eDP+HDMI connected 2. Boot it 3. Kill the display manager so fbcon shows up 4. sudo ./kms_frontbuffer_tracking --show-hidden --run-subtest nop-1p-rte 5. If the machine is not frozen (50% chance), wait a few seconds and then goto step 4. The machine freeze is a complete freeze. You can even see a red led lighting up on the motherboard. Notice that even though the subtest we're running only lights up a single pipe, the bug only happens if I have eDP+HDMI connected. Now, if you open kms_frontbuffer_tracking.c, go to function rte_subtest() and then uncomment the call to set_sprite_for_test(), then you won't be able to freeze the machine. Notice that I've been experiencing different types of dual-monitor problems on SKL even since. Previously, I couldn't even boot with eDP+HDMI because the machine would freeze while loading the driver. Since some watermarks patches solved this issue, I believe this bug may also have some relationship with watermarks. It is also worth mentioning that just booting my SKL machine with the two monitors connected already gives me dmesg errors, such as WARN_ON(!wm_changed).
Paulo - do you still see this with the latest kernel, preferable compiled from drm-tip? On 4.10.0-drm-tip-qa-ww9-commit-7f38bb9+ (executed 9.3.2017) with SKL-1-NUC6i5SYB the case mentioned is having pass as verdict. The difference here is that on our SKL we have Connector info -------------- connector 48: type DP-1, status: disconnected connector 52: type HDMI-A-1, status: connected connector 55: type DP-2, status: connected connector 59: type HDMI-A-2, status: disconnected Two displays, yes, but now eDP (assumable).
(In reply to Jari Tahvanainen from comment #1) > Paulo - do you still see this with the latest kernel, preferable compiled > from drm-tip? On 4.10.0-drm-tip-qa-ww9-commit-7f38bb9+ (executed 9.3.2017) > with SKL-1-NUC6i5SYB the case mentioned is having pass as verdict. The > difference here is that on our SKL we have Connector info > -------------- > connector 48: type DP-1, status: disconnected > connector 52: type HDMI-A-1, status: connected > connector 55: type DP-2, status: connected > connector 59: type HDMI-A-2, status: disconnected > Two displays, yes, but now eDP (assumable). Mistype: Two displays, yes, but NOT eDP (assumable).
I can confirm this is gone with a recent drm-intel-nightly in the machine I used to reproduce it. I suppose this probably got fixed last year when we addressed the watermarks patches. I'll mark this as fixed.
this issue is confirmed fixed...closing
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.