According to Documentation/mm/transhuge.rst, MADV_COLLAPSE is expected to ignore any THP or mTHP interface settings. However, after commit 6beeab870e70 ("mm: shmem: move shmem_huge_global_enabled() into shmem_allowable_huge_orders()"), performing MADV_COLLAPSE on shmem will depend on /sys/.../hugepages-2048kB/shmem_enabled being set to "inherit" (although that is the default), which can cause MADV_COLLAPSE to fail unexpectedly. Fix this by returning the result of shmem_huge_global_enabled() directly when MADV_COLLAPSE is requested, allowing PMD-order collapse while ignoring shmem THP/mTHP settings. Fixes: 6beeab870e70 ("mm: shmem: move shmem_huge_global_enabled() into shmem_allowable_huge_orders()") Reported-by: Kiryl Shutsemau (Meta) Closes: https://lore.kernel.org/all/20260908125105.1510704-7-kirill@shutemov.name/ Signed-off-by: Baolin Wang --- Hi Kiryl, I did not update the selftests, as I hope this fix will be merged first. After this fix has landed, I can send a patch to update the selftests. --- mm/shmem.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/mm/shmem.c b/mm/shmem.c index c92ed17dbc4a..b572c60f2af8 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -2049,8 +2049,11 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode, global_orders = shmem_huge_global_enabled(inode, index, write_end, shmem_huge_force, vma, vm_flags); - /* Tmpfs huge pages allocation */ - if (!vma || !vma_is_anon_shmem(vma)) + /* + * Tmpfs huge pages allocation or forced collapse ignores + * sysfs configs. + */ + if (!vma || !vma_is_anon_shmem(vma) || shmem_huge_force) return global_orders; /* -- 2.47.3