Hi, As advised by mszyprow on IRC, I am coming here to report a graphics corruption issue that happens on the Exynos 4412 SoC whenever we use devfreq in conjunction with drm overlay planes (DRM_PLANE_TYPE_OVERLAY). We are working on a fork of LineageOS 16 (based on AOSP 9) that uses drm_hwcomposer attached to an exynos card0 node. Whenever we enabled overlay planes we would get the graphics corruptions that can be seen at https://gitlab.freedesktop.org/drm-hwcomposer/drm-hwcomposer/issues/24 With the help of ahajda on #linux-exynos the root cause of this issue was traced into devfreq. devfreq was enabled on our 5.2 kernel branch as follows: CONFIG_PM_DEVFREQ=y CONFIG_DEVFREQ_GOV_PERFORMANCE=y CONFIG_DEVFREQ_GOV_POWERSAVE=y CONFIG_DEVFREQ_GOV_USERSPACE=y CONFIG_ARM_EXYNOS_BUS_DEVFREQ=y CONFIG_DEVFREQ_EVENT_EXYNOS_NOCP=y Removing these configs (i.e. disabling devfreq) got rid of all corruption issues. ahajda provided a tentative explanation for this: > There are multiple issues with devfreq and clocks, devfreq too > aggresively lowers display clock frequencies, as a result DMA > transfers of overlays are too slow and result in screen corruption We are reporting here in case someone can take this task, as we lack the skill and knowledge to fix it ourselves. We were informed on IRC that the fix could take the form of a patch similar to what mszyprow did for the mixer/hdmi path ( https://patchwork.kernel.org/patch/10861757/ ) but this time targeting the fimd/lcd panel path. If needed we can gather more information and provide all assistance you need for testing. Some logs are already available at https://gitlab.freedesktop.org/drm-hwcomposer/drm-hwcomposer/issues/24 The relevant IRC discussion is archived at https://freenode.irclog.whitequark.org/linux-exynos/2019-07-18 Regards, David
