Re: [PATCH v5 5/5] selftests/mm: add UFFDIO_MOVE ioctl test

[David Hildenbrand]

On 02.12.23 09:04, Ryan Roberts wrote:
> On 01/12/2023 20:47, David Hildenbrand wrote:
> > On 01.12.23 10:29, Ryan Roberts wrote:
> > > On 21/11/2023 17:16, Suren Baghdasaryan wrote:
> > > > Add tests for new UFFDIO_MOVE ioctl which uses uffd to move source
> > > > into destination buffer while checking the contents of both after
> > > > the move. After the operation the content of the destination buffer
> > > > should match the original source buffer's content while the source
> > > > buffer should be zeroed. Separate tests are designed for PMD aligned and
> > > > unaligned cases because they utilize different code paths in the kernel.
> > > >
> > > > Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> > > > ---
> > > >    tools/testing/selftests/mm/uffd-common.c     |  24 +++
> > > >    tools/testing/selftests/mm/uffd-common.h     |   1 +
> > > >    tools/testing/selftests/mm/uffd-unit-tests.c | 189 +++++++++++++++++++
> > > >    3 files changed, 214 insertions(+)
> > > >
> > > > diff --git a/tools/testing/selftests/mm/uffd-common.c
> > > > b/tools/testing/selftests/mm/uffd-common.c
> > > > index fb3bbc77fd00..b0ac0ec2356d 100644
> > > > --- a/tools/testing/selftests/mm/uffd-common.c
> > > > +++ b/tools/testing/selftests/mm/uffd-common.c
> > > > @@ -631,6 +631,30 @@ int copy_page(int ufd, unsigned long offset, bool wp)
> > > >        return __copy_page(ufd, offset, false, wp);
> > > >    }
> > > >    +int move_page(int ufd, unsigned long offset, unsigned long len)
> > > > +{
> > > > +    struct uffdio_move uffdio_move;
> > > > +
> > > > +    if (offset + len > nr_pages * page_size)
> > > > +        err("unexpected offset %lu and length %lu\n", offset, len);
> > > > +    uffdio_move.dst = (unsigned long) area_dst + offset;
> > > > +    uffdio_move.src = (unsigned long) area_src + offset;
> > > > +    uffdio_move.len = len;
> > > > +    uffdio_move.mode = UFFDIO_MOVE_MODE_ALLOW_SRC_HOLES;
> > > > +    uffdio_move.move = 0;
> > > > +    if (ioctl(ufd, UFFDIO_MOVE, &uffdio_move)) {
> > > > +        /* real retval in uffdio_move.move */
> > > > +        if (uffdio_move.move != -EEXIST)
> > > > +            err("UFFDIO_MOVE error: %"PRId64,
> > > > +                (int64_t)uffdio_move.move);
> > >
> > > Hi Suren,
> > >
> > > FYI this error is triggering in mm-unstable (715b67adf4c8):
> > >
> > > Testing move-pmd on anon... ERROR: UFFDIO_MOVE error: -16 (errno=16,
> > > @uffd-common.c:648)
> > >
> > > I'm running in a VM on Apple M2 (arm64). I haven't debugged any further, but
> > > happy to go deeper if you can direct.
> >
> > Does it trigger reliably? Which pagesize is that kernel using?
>
> Yep, although very occasionally it fails with EAGAIN. 4K kernel; see other email
> for full config.
>
> > I can spot that uffd_move_pmd_test()/uffd_move_pmd_handle_fault() uses
> > default_huge_page_size(), which reads the default hugetlb size.
>
> My kernel command line is explicitly seting the default huge page size to 2M.

Okay, so that likely won't affect it.

I can only guess that it has to do with the alignment of the virtual 
area we are testing with, and that we do seem to get more odd patterns 
on arm64.

uffd_move_test_common() is a bit more elaborate, but if we aligned the 
src+start area up, surely "step_count" cannot be left unmodified?

So assuming we get either an unaligned source or an unaligned dst from 
mmap(), I am not convinced that we won't be moving areas that are not 
necessarily fully backed by PMDs and maybe don't even fall into the VMA 
of interest?

Not sure if that could trigger the THP splitting issue, though.

But I just quickly scanned that test setup, could be I am missing 
something. It might make sense to just print the mmap'ed range and the 
actual ranges we are trying to move. Maybe something "obvious" can be 
observed.

--
Cheers,

David / dhildenb