I believe the mutex type patch has its own value in protecting memory_failure from other inherent races, e.g., races around split_huge_page where concurrent MCE happens to different 4k pages under the same THP.[1] This realistically can happen given the physical locality clustering effect of memory errors.
Thanks, -Jue [1] The split fails due to a page reference taken by other concurrent calling into memory_failure on the same THP.
