Linux is a highly configurable operating-system kernel which has been widely studied in the context of software product lines over the past years. Understanding the challenges and perils of evolving and maintaining feature models of the size of Linux is crucial to provide the right tools for development today and to direct future research. Unfortunately, previous studies show contradictory observations when analyzing the evolution of Linux feature models. We explain how peculiarities of the feature models of the Linux kernel lead to those differing observations, and show how the results can be re-aligned. Moreover, our findings also demonstrate that symbolic differencing on feature models used by researchers so far has limited value, depending on the use case. We show how the limitations can be addressed by means of semantic differencing, and ironically invalidate the results we sought to re-align.