On Sat, Jul 11, 2020 at 11:12 AM Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > The fact that it seems to happen with > > https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/mem/thp/thp01.c > > makes me think it's somehow related to THP mappings, but I don't see > why those would matter. All the same pmd freeing should still have > happened, afaik. No, I think the only reason it happens with that test-case is simply because that test-case tests many different argument sizes, and as part of that it ends up then hitting the "stack is exactly one pmd in size" case when the stack alignment is right too. So the THP part is almost certainly a red herring. Maybe. Considering that I don't see what's wrong, anything is possible. Linus
