On 4/8/26 2:02 PM, Andrew Morton wrote:
On Wed, 8 Apr 2026 13:35:42 -0700 [email protected] wrote:
On 4/3/26 12:31 PM, Andrew Morton wrote:
On Thu, 2 Apr 2026 16:59:33 -0700 Anthony Yznaga <[email protected]>
wrote:
For configs that support MAP_DROPPABLE verify that a mapping created
with MAP_DROPPABLE cannot be locked via mlock(), and that it will not
be locked if it's created after mlockall(MCL_FUTURE).
There are a few queries from the AI reviewbot;
https://sashiko.dev/#/patchset/[email protected]
Interesting. Of the two issues, one is certainly legit. I need to add an
munlockall() on early return from test_mlockall_future_droppable().
Cool.
For the other, the question posed was whether the tests should handle
possibly being run on an older kernel that doesn't implement
MAP_DROPPABLE. It seems to me to that a selftest should not be expected
to work (or even necessarily compile) on kernels older than when the
selftest was introduced, but I don't want to assume.
I don't know that there's any policy on that. My attitude is that
selftests are not intended to be forward- or backward-compatible.
That's why we ship them with the kernel source!
If we get a selftests fixup then I do like to backport that into
earlier kernels if appropriate, to keep those in good shape. And that
has the effect of reducing people's motivation to run a later kernel's
selftests on their current kernel.
That makes sense. It's trivial to skip the tests if MAP_DROPPABLE is not
defined so I'll do that.
