Thanks! This looks like an upstream regression specific to 3168 model, which unfortunately we backported as the fix as well: https://bugzilla.kernel.org/show_bug.cgi?id=206329 I'm going to provide you a test kernel with the suggested upstream fix.