Re: [PATCH v2 2/2] kselftest: mm: replace usage of /proc/self/smaps for check_huge_xxx() helper

[Yeoreum Yun <yeoreum.yun@arm.com>]

On Thu, Aug 27, 2026 at 12:56:20PM +0200, David Hildenbrand (Arm) wrote:
> On 8/27/26 10:28, Baolin Wang wrote:
> > 
> > 
> > On 8/26/26 8:24 PM, Yeoreum Yun wrote:
> >> Since glibc commit 321e1fc73f (“malloc: Enable 2MB THP by default on AArch64”),
> >> glibc may call madvise(MADV_HUGEPAGE) for sufficiently large allocations
> >> made by memalign().
> >>
> >> The underlying VMA may start at a different address from the aligned
> >> address returned by memalign(). Furthermore, a subsequent
> >> madvise(MADV_HUGEPAGE) call does not split the VMA because the flag is
> >> already set.
> >>
> >> This causes split_huge_page_test to fail because the check_huge_xxx()
> >> helpers incorrectly require the address returned by memalign() to
> >> match the VMA start address reported in /proc/self/smaps.
> >>
> >> Fix this by using /proc/self/pagemap and /proc/kpageflags instead of
> >> /proc/self/smaps to detect huge pages.
> >>
> >> Reported-by: David Hildenbrand (Arm) <david@kernel.org>
> >> Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
> >> ---
> >>   tools/testing/selftests/mm/vm_util.c | 130 ++++++++++++++++++++---------------
> >>   tools/testing/selftests/mm/vm_util.h |   1 +
> >>   2 files changed, 77 insertions(+), 54 deletions(-)
> >>
> >> diff --git a/tools/testing/selftests/mm/vm_util.c b/tools/testing/selftests/
> >> mm/vm_util.c
> >> index 4821a3563036..1d0959b3b9e8 100644
> >> --- a/tools/testing/selftests/mm/vm_util.c
> >> +++ b/tools/testing/selftests/mm/vm_util.c
> >> @@ -351,31 +351,13 @@ char *__get_smap_entry(void *addr, const char *pattern,
> >> char *buf, size_t len)
> >>       return entry;
> >>   }
> >>   -static bool __check_pmd_huge(void *addr, char *pattern, int nr_hpages,
> >> -          uint64_t hpage_size)
> >> -{
> >> -    char buffer[MAX_LINE_LENGTH];
> >> -    uint64_t thp = -1;
> >> -    char *entry;
> >> -
> >> -    entry = __get_smap_entry(addr, pattern, buffer, sizeof(buffer));
> >> -    if (!entry)
> >> -        goto err_out;
> >> -
> >> -    if (sscanf(entry, "%9" SCNu64 " kB", &thp) != 1)
> >> -        ksft_exit_fail_msg("Reading smap error\n");
> >> -
> >> -err_out:
> >> -    return thp == (nr_hpages * (hpage_size >> 10));
> >> -}
> >> -
> >> -static bool check_large_folios(void *addr, size_t len, int nr_hpages,
> >> -        uint64_t hpage_size)
> >> +static bool check_large_folios(int pagemap_fd, int kpageflags_fd,
> >> +                   void *addr, size_t len, int nr_hpages,
> >> +                   uint64_t hpage_size)
> >>   {
> >>       int order = 0, pagesize = getpagesize();
> >>       unsigned int nr_pages = hpage_size / pagesize;
> >>       int orders[MAX_NR_ORDERS], status;
> >> -    int pagemap_fd, kpageflags_fd;
> >>       bool ret = false;
> >>         if (!nr_pages)
> >> @@ -386,15 +368,6 @@ static bool check_large_folios(void *addr, size_t len,
> >> int nr_hpages,
> >>           ksft_exit_fail_msg("invalid order\n");
> >>         memset(orders, 0, sizeof(int) * MAX_NR_ORDERS);
> >> -    pagemap_fd = open(PAGEMAP_PATH, O_RDONLY);
> >> -    if (pagemap_fd == -1)
> >> -        ksft_exit_fail_msg("read pagemap fail\n");
> >> -
> >> -    kpageflags_fd = open(KPAGEFLAGS_PATH, O_RDONLY);
> >> -    if (kpageflags_fd == -1) {
> >> -        close(pagemap_fd);
> >> -        ksft_exit_fail_msg("read kpageflags fail\n");
> >> -    }
> >>         status = gather_folio_orders(addr, len, pagemap_fd,
> >>               kpageflags_fd, orders, MAX_NR_ORDERS);
> >> @@ -405,48 +378,97 @@ static bool check_large_folios(void *addr, size_t len,
> >> int nr_hpages,
> >>           ret = true;
> >>     out:
> >> -    close(pagemap_fd);
> >> -    close(kpageflags_fd);
> >>       return ret;
> >>   }
> >>   -bool check_huge_anon(void *addr, size_t len, int nr_hpages, uint64_t
> >> hpage_size)
> >> +enum check_huge_type {
> >> +    CHECK_HUGE_ANON,
> >> +    CHECK_HUGE_FILE,
> >> +    CHECK_HUGE_SHMEM,
> >> +};
> >> +
> >> +static bool __check_pmd_huge(void *addr, size_t len, int nr_hpages,
> >> +                 uint64_t hpage_size, enum check_huge_type type)
> > 
> > The original __check_pmd_huge() is only for PMD-sized large folios, but now it
> > not only checks PMD-sized large folios but also mTHP large folios, which I find
> > confusing. Please keep its original semantics, and only check PMD-sized large
> > folios.
> > 
> >>   {
> >> -    uint64_t pmd_pagesize = read_pmd_pagesize();
> >> +    int pagemap_fd, kpageflags_fd;
> >> +    uint64_t pmd_pagesize, granule;
> >> +    uint64_t categories, kpf;
> >> +    unsigned long pfn;
> >> +    bool check_large, huge_mapped;
> >> +    char *start = addr;
> >> +    char *end = start + len;
> >>   +    pmd_pagesize = read_pmd_pagesize();
> >>       if (!pmd_pagesize)
> >>           ksft_exit_fail_msg("reading PMD pagesize failed\n");
> >>   -    if (hpage_size == pmd_pagesize)
> >> -        return __check_pmd_huge(addr, "AnonHugePages: ", nr_hpages, hpage_size);
> >> +    if (nr_hpages > 0) {
> >> +        check_large = true;
> >> +        granule = hpage_size;
> >> +    } else {
> >> +        check_large = false;
> >> +        granule = psize();
> >> +    }
> > 
> > This is incorrect for the mTHP large folio check. I already hit a selftest
> > failure. Please test your patches before sending them out.
> > 
> > [root@]./khugepaged -c 4 mthp_khugepaged:anon
> > TAP version 13
> > # Save THP and khugepaged settings... OK
> > 1..4
> > # Allocate huge page on fault... OK
> > # Split huge PMD on MADV_DONTNEED... OK
> > ok 1 allocate on fault and split
> > #
> > # Run test: collapse_full (mthp_khugepaged:anon)
> > # Collapse multiple fully populated PTE table.... OK
> > ok 2 collapse_full
> > #
> > # Run test: collapse_empty (mthp_khugepaged:anon)
> > # Do not collapse empty PTE table.... OK
> > ok 3 collapse_empty
> > #
> > # Run test: collapse_single_mthp (mthp_khugepaged:anon)
> > # Collapse PTE table with half PTE entries present.... Fail
> > not ok 4 collapse_single_mthp
> > # Totals: pass:3 fail:1 xfail:0 xpass:0 skip:0 error:0
> 
> FWIW, the CI flags this as well:
> 
> https://github.com/linux-mm/linux-mm/actions/runs/33028595020/job/98375618491

Yeap. I've overlooked that case and here is the fix:
  - https://lore.kernel.org/all/apAVAvfxZSeEvf50@e129823.arm.com/
> 
> -- 
> Cheers,
> 
> David

-- 
Sincerely,
Yeoreum Yun