Re: [PATCH 1/2] selftests: khugepaged: fix the shmem collapse failure

[Baolin Wang]

On 2025/6/12 13:10, Dev Jain wrote:
> On 12/06/25 9:24 am, Baolin Wang wrote:
> > When running the khugepaged selftest for shmem (./khugepaged all:shmem),
> > I encountered the following test failures:
> > "
> > Run test: collapse_full (khugepaged:shmem)
> > Collapse multiple fully populated PTE table.... Fail
> > ...
> > Run test: collapse_single_pte_entry (khugepaged:shmem)
> > Collapse PTE table with single PTE entry present.... Fail
> > ...
> > Run test: collapse_full_of_compound (khugepaged:shmem)
> > Allocate huge page... OK
> > Split huge page leaving single PTE page table full of compound 
> > pages... OK
> > Collapse PTE table full of compound pages.... Fail
> > "
> >
> > The reason for the failure is that, it will set MADV_NOHUGEPAGE to 
> > prevent
> > khugepaged from continuing to scan shmem VMA after khugepaged finishes
> > scanning in the wait_for_scan() function. Moreover, shmem requires a 
> > refault
> > to establish PMD mappings.
> >
> > However, after commit 2b0f922323cc, PMD mappings are prevented if the 
> > VMA is
> > set with MADV_NOHUGEPAGE flag, so shmem cannot establish PMD mappings 
> > during
> > refault.
> >
> > To fix this issue, we can set the MADV_NOHUGEPAGE flag after the shmem 
> > refault.
> > With this fix, the shmem test case passes.
> >
> > Fixes: 2b0f922323cc ("mm: don't install PMD mappings when THPs are 
> > disabled by the hw/process/vma")
> > Signed-off-by: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>
> > ---
> >   tools/testing/selftests/mm/khugepaged.c | 3 +--
> >   1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/tools/testing/selftests/mm/khugepaged.c 
> > b/tools/testing/selftests/mm/khugepaged.c
> > index 8a4d34cce36b..d462f62d8116 100644
> > --- a/tools/testing/selftests/mm/khugepaged.c
> > +++ b/tools/testing/selftests/mm/khugepaged.c
> > @@ -561,8 +561,6 @@ static bool wait_for_scan(const char *msg, char 
> > *p, int nr_hpages,
> >           usleep(TICK);
> >       }
> > -    madvise(p, nr_hpages * hpage_pmd_size, MADV_NOHUGEPAGE);
> > -
> >       return timeout == -1;
> >   }
> > @@ -585,6 +583,7 @@ static void khugepaged_collapse(const char *msg, 
> > char *p, int nr_hpages,
> >       if (ops != &__anon_ops)
> >           ops->fault(p, 0, nr_hpages * hpage_pmd_size);
> > +    madvise(p, nr_hpages * hpage_pmd_size, MADV_NOHUGEPAGE);
> >       if (ops->check_huge(p, expect ? nr_hpages : 0))
> >           success("OK");
> >       else
>
> The idea looks sane to me, but do we need to add the madvise call to
> madvise_retracted_page_tables() too, since that also calls wait_for_scan()?

Yes, I also realized this after sending the patches:) Thanks.

To keep the original logic:

diff --git a/tools/testing/selftests/mm/khugepaged.c 
b/tools/testing/selftests/mm/khugepaged.c
index d462f62d8116..3452763e2fe3 100644
--- a/tools/testing/selftests/mm/khugepaged.c
+++ b/tools/testing/selftests/mm/khugepaged.c
@@ -1074,6 +1074,7 @@ static void madvise_retracted_page_tables(struct 
collapse_context *c,
                return;
        }
        success("OK");
+       madvise(p, nr_hpages * hpage_pmd_size, MADV_NOHUGEPAGE);
        c->collapse("Install huge PMD from page cache", p, nr_hpages, ops,
                    true);
        validate_memory(p, 0, size);