Ok, I'll just remove that HAVE_FAST_GUP - I was expecting it to come through David, but I was also really hoping we'd have a guess as to why it happens on some hardware and configurations but not others.
It works, thank you - the mutex hang regression in util-linux test suite no longer triggers and we can move forward with sparc64 kernels in debian-ports and also re-check any other regressions found meanwhile. I will recheck my sparc64's and report any surviving regressions if any. -- Meelis Roos <mroos@xxxxxxxx>