Hi,
thanks a lot for the review, Andrzej!
Andrzej Hajda:
+static const struct backlight_ops s6e8aa0_backlight_ops = {
- .update_status = s6e8aa0_set_brightness,
This is racy, update_status can be called in any time between probe and remove, particularly:
a) before panel enable,
b) during panel enable,
c) when panel is enabled,
d) during panel disable,
e) after panel disable,
b and d are racy for sure - backlight and drm callbacks are async.
IMO the best solution would be to register backlight after attaching panel to drm, but for this drm_panel_funcs should have attach/detach callbacks (like drm_bridge_funcs),
then update_status callback should take some drm_connector lock to synchronize with drm, and write to hw only when pipe is enabled.
I will start reading the kernel DRM KMS documentation in order to learn how the attach callback would fit in the picture and do what you suggest but it might take a few weeks or months.
Also, as a lot of this code was mimicked from drivers/gpu/drm/panel/panel-samsung-s6e63m0.c it looks like that is also having this race condition. Unfortunately, I don't think I have the s6e63m0 panel so for that I probably cannot fix this issue.
Joonas