Hi Daniel, all,
So, I was playing around with arcpgu and I found a bug at module unloading. I corrected it and tested (using a 4.12 kernel) and I got a nice WARNING from DRM core at unloading (instead of the old NULL pointer I got :D). I debugged it and found out that an empty modesetting (i.e. a mode with no clock) was being set by the core and validated with crtc->atomic_check() callback. At arcpgu we are expecting the clock value to be valid and > 0 so we were rejecting the empty modesetting which caused the core to not set the final mode before unloading.
I guess then it's expectable that atomic_check()/mode_valid() callbacks can be called with an empty mode, right? Is this documented anywhere?
Also, do we really need to call atomic check in this case? I mean, its an empty modesetting and the mode is not valid so it should be rejected, no?
Also notice that this error doesn't happen with 4.13 because we switched to crtc->mode_valid() callback which instead is only called if a full crtc/encoder/connector pipeline is set, which in the case of unloading is not.
Best regards, Jose Miguel Abreu