On Sun Aug 2, 2026 at 3:52 PM EDT, Kiryl Shutsemau wrote: > From: "Kiryl Shutsemau (Meta)" <[email protected]> > > The page cache caps folio order at MAX_PAGECACHE_ORDER, and > xas_split_alloc() puts that cap below the PMD order where a PMD is 512M -- > arm64 with 64K base pages, as include/linux/pagemap.h says outright. > shmem_huge_global_enabled() then offers no PMD order at all, so > MADV_COLLAPSE of a shmem range answers -EINVAL and khugepaged passes over > it.
IIRC, after READ_ONLY_THP_FOR_FS is removed, all pagecache folios are split using non uniform split, xas_try_split(), so does shmem (except shmem in swapcache not splittable). In theory, we can get rid of the cap, since xas_try_split() does not split more than one level like one can try to make xas_split_alloc() split more than two level (e.g., 512MB to 64KB on arm64 with 64KB base page). > > The shmem cases nonetheless ask for a PMD-sized shmem folio, so on such a > configuration four of them fail and the run bails out in the middle: > > not ok 2 collapse_full not ok 4 collapse_single_pte_entry # Allocate huge > page...Bail out! madvise(MADV_COLLAPSE): Invalid argument (22) > > That is the kernel declining something it deliberately does not support, > not a collapse defect. Skip those cases where the PMD order is not a shmem > order, which thp_shmem_supported_orders() already reports -- it reads the > same per-size shmem_enabled controls the kernel only publishes for orders > the page cache can hold. A tmpfs-backed file argument is skipped on the > same grounds, and a run left with nothing to collapse into skips outright. > > Anonymous collapse is unaffected: its orders are not capped this way, and > the anonymous cases pass at a 512M PMD. > > Assisted-by: Claude-Code:claude-opus-5 > Signed-off-by: Kiryl Shutsemau (Meta) <[email protected]> > --- > tools/testing/selftests/mm/khugepaged.c | 21 +++++++++++++++++++++ > 1 file changed, 21 insertions(+) > > diff --git a/tools/testing/selftests/mm/khugepaged.c > b/tools/testing/selftests/mm/khugepaged.c > index 9213ce1658d0..c5a3c3922581 100644 > --- a/tools/testing/selftests/mm/khugepaged.c > +++ b/tools/testing/selftests/mm/khugepaged.c > @@ -1584,6 +1584,27 @@ int main(int argc, char **argv) > hpage_pmd_nr = hpage_pmd_size / page_size; > hpage_pmd_order = __builtin_ctz(hpage_pmd_nr); > > + /* > + * The page cache caps folio order at MAX_PAGECACHE_ORDER, which > + * xas_split_alloc() puts below the PMD order on arm64 with 64K pages. > + * A PMD-sized page cache folio is then impossible, so the kernel > + * refuses these collapses by design and there is nothing to test. > + */ > + if (!(thp_shmem_supported_orders() & (1UL << hpage_pmd_order))) { > + if (shmem_ops) { > + ksft_print_msg("no PMD-order page cache folio: skipping > shmem\n"); > + shmem_ops = NULL; > + } > + if (finfo.type == VMA_SHMEM && read_only_file_ops) { > + ksft_print_msg("no PMD-order page cache folio: skipping > tmpfs file\n"); > + read_only_file_ops = NULL; > + read_write_file_read_ops = NULL; > + read_write_file_write_ops = NULL; > + } > + if (!anon_ops && !shmem_ops && !read_only_file_ops) > + ksft_exit_skip("Nothing left to collapse into\n"); > + } > + > if (anon_target_order && > !(thp_supported_orders() & (1UL << anon_target_order))) > ksft_exit_skip("Order %d is not a supported anon THP order\n", -- Best Regards, Yan, Zi
