The another S3 regression is tracked in boo#1030698. BTW, I got a Dell SKL machine and I could reproduce the problem locally, so it's easier for me now to debug :)