On Wed, 2023-12-06 at 11:57 -0500, Andrei Matei wrote: [...] > > I think we also need a selftest, at-least for patch #1. > > Yeah... I was wondering in a message on v1 [1] if a test like what I had > prototyped there was warranted. I personally still think it's not because of > how many other variations of variable-offset stack access tests we already > have, and how random the zero-sized read case is (even though we had a bug in > there) - so protecting against this particular regression didn't seem worth it > to me. I'm now including the test in v4 but if you change your mind about it > when you see it in context, let me know and I'll take it out. > > [1] https://lore.kernel.org/bpf/CABWLsevk47Xa1a+h0UK--94zEuxScrmyt0-D8YShq1UgvVvf5g@xxxxxxxxxxxxxx/ The reproducer is a small and non-contrived program, so I think there is no harm in adding it.
