hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/IAEOV4 CVE: NA
--------------------------------
When thp_exec_enabled is set to 3, try PMD mapping THP first, if failed, then try mTHP like cont-64K on ARM64. The previous implementation use hugepage_madvise() in try_enable_file_exec_thp() and it will always set VM_HUGEPAGE to vm_flags, so we lost this semantic, let's check PMD_ORDER by thp_vma_allowable_order() to correct the semantic.
And now the khugepaged_enter_vma() step in hugepage_madvise() is missing, this is OK because the process is already entered in khugepaged after this commit 613bec092fe7 ("mm: mmap: register suitable readonly file vmas for khugepaged").
Signed-off-by: Nanyong Sun sunnanyong@huawei.com --- mm/filemap.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/mm/filemap.c b/mm/filemap.c index d3c813429bf2..4dd8fe6ffa05 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -3159,8 +3159,12 @@ static inline void try_enable_file_exec_thp(struct vm_area_struct *vma, if (file->f_op->get_unmapped_area != thp_get_unmapped_area) return;
- if (file_exec_thp_enabled()) - hugepage_madvise(vma, vm_flags, MADV_HUGEPAGE); + if (!file_exec_thp_enabled()) + return; + + if (thp_vma_allowable_order(vma, *vm_flags, TVA_ENFORCE_SYSFS, + PMD_ORDER)) + *vm_flags |= VM_HUGEPAGE; }
static inline bool file_exec_can_enable_mthp(struct address_space *mapping,