Created attachment 131969 [details] dmesg log: boot with eDP; HDMI hotplugged later Some dual screen setups are currently broken. All the scanlines below the mouse cursor are wrong, sometimes containing a shade of gray and others repeating the image above it. I've seen this in APL and GLK using DSI+HDMI, eDP+HDMI and DSI+DP. In the DP case, however, the screen goes blank and reports lack of signal. git bisect points at commit bb9d85f6e9de8fef5236c076530eab67a2f2431b Author: Kumar, Mahesh <mahesh1.kumar@intel.com> Date: Thu Jun 1 11:29:18 2017 +0530 drm/i915/skl: New ddb allocation algorithm
Created attachment 131984 [details] [review] Don't fix cursor DDB Hi Ander, As per our debugging, In new algorithm minimum WM requirement for Cursor plane is increasing & it's more than cursor allocated DDB (8-blocks). Because of which Cursor WM are getting disabled in system. I created one RFC patch for not hard-coding cursor DDB to 8 in multi-pipe scenario, instead select a value from a particular range. This way we avoid re computing complete pipe's WM in each cursor update as well. Please let me know if it helps.
Changing the status to NEEDINFO, once you provide more information, please change the tag to REOPEN. Thank you.
I tested the patch and it did fix the corrupted scanlines when moving the mouse around. My setup was APL with DSI+DP connected.
I'm not sure what is the next step here. The revert was merged, so should this be closed? In any case, the patch fixed the issue in my GLK with eDP+HDMI too, tested on top of a drm-tip prior to the revert. We definitely need to improve the test coverage to test dual screen. This is a very serious regression that slipped through. It would be good to have an igt test that exercises different configurations with varying DDB requirements.
(In reply to Ander Conselvan de Oliveira from comment #4) > I'm not sure what is the next step here. The revert was merged, so should > this be closed? > > In any case, the patch fixed the issue in my GLK with eDP+HDMI too, tested > on top of a drm-tip prior to the revert. > > We definitely need to improve the test coverage to test dual screen. This is > a very serious regression that slipped through. It would be good to have an > igt test that exercises different configurations with varying DDB > requirements. Closing bug since is reported as fixed. If there is any new problem concerning this case, please change the tag to "REOPEN". Thank you.
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.