On Wed, Sep 28, 2022 at 04:58:53PM +0000, Sean Christopherson wrote:
> On Tue, Sep 27, 2022, Ricardo Koller wrote:
> > On Tue, Sep 27, 2022 at 10:06:23PM +0000, Sean Christopherson wrote:
> > > On Mon, Sep 26, 2022, Ricardo Koller wrote:
> > > > On Thu, Sep 22, 2022 at 07:32:42PM +0000, Sean Christopherson wrote:
> > > > > On Thu, Sep 22, 2022, Ricardo Koller wrote:
> > > > > > + void *hva = (void *)region->region.userspace_addr;
> > > > > > + uint64_t paging_size = region->region.memory_size;
> > > > > > + int ret, fd = region->fd;
> > > > > > +
> > > > > > + if (fd != -1) {
> > > > > > + ret = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
> > > > > > + 0, paging_size);
> > > > > > + TEST_ASSERT(ret == 0, "fallocate failed, errno: %d\n", errno);
> > > > > > + } else {
> > > > > > + if (is_backing_src_hugetlb(region->backing_src_type))
> > > > > > + return false;
> > > > >
> > > > > Why is hugetlb disallowed? I thought anon hugetlb supports MADV_DONTNEED?
> > > > >
> > > >
> > > > It fails with EINVAL (only tried on arm) for both the PAGE_SIZE and the huge
> > > > page size. And note that the address is aligned as well.
> > > >
> > > > madvise(0xffffb7c00000, 2097152, MADV_DONTNEED) = -1 EINVAL (Invalid argument)
> > > > ^^^^^^^^^^^^^^ ^^^^^^^
> > > > 2M aligned 2M (hugepage size)
> > > >
> > > > madvise(0xffff9e800000, 4096, MADV_DONTNEED) = -1 EINVAL (Invalid argument)
> > > > ^^^^
> > > > PAGE_SIZE
> > >
> > > I think this needs to be root caused before merging. Unless I'm getting turned
> > > around, MADV_DONTEED should work, i.e. there is a test bug lurking somewhere.
> >
> > Turns out that the failure is documented. Found this in the madvise manpage:
> >
> > MADV_DONTNEED cannot be applied to locked pages, Huge TLB pages, or VM_PFNMAP pages.
>
> The manpages are stale:
>
> c4b6cb884011 ("selftests/vm: add hugetlb madvise MADV_DONTNEED MADV_REMOVE test")
> 90e7e7f5ef3f ("mm: enable MADV_DONTNEED for hugetlb mappings")
>
> The tools/testing/selftests/vm/hugetlb-madvise.c selftest effectively tests what
> is being done here, so _something_ is broken.
Thanks for the pointers. I was using old kernels (~5.15) for these
latest tests. Testing on a 6.0-rc3 kernel fixed things: now able to
madvise(MADV_DONTNEED) on anon-hugetlb from the selftest (arm).
Will remove the check (for skppping the test) in v9.
Thanks!
Ricardo
