On 28.08.24 14:30, Cyril Hrubis wrote:
Hi!
If we are going to stick to ENOENT for page that wasn't faulted in the
kernel from now on we should stick to it in the test as well.
Right, it will make kernels >= 4.3 fail, though, until this series is
upstream. I mean, it highlights a BUG, but we had a similar condition
with the zeropage and worked around it in the test to keep it passing.
What we do in this case is that you are free to send a patch, we will do
review but the final merge will happen once the code has been released
in the upstream kernel.
Yes, let me send something out for discussion.
Also I think there is a third case that we do not cover either, what
happens when we pass an address that is not mapped at all, e.g. NULL? Do
we get EFAULT as well?
Yes, that's documented as EFAULT and should behave that way. I can
extend the test to handle that as well.
Ideally the test should be ported to the new test API as well, but I can
do that later on the top of your work.
I tried, and it all looked easy, until I realized that these tests use a
share code-base:
testcases/kernel/syscalls/move_pages/move_pages_support.c
That is also written using the old API. I assume mixing APIs might not
work as expected ...
--
Cheers,
David / dhildenb
