From: Mike Kravetz <mike.kravetz@xxxxxxxxxx> commit dff11abe280b47c21b804a8ace318e0638bb9a49 upstream. When fixing an issue with PMD sharing and migration, it was discovered via code inspection that other callers of huge_pmd_unshare potentially have an issue with cache and tlb flushing. Use the routine adjust_range_if_pmd_sharing_possible() to calculate worst case ranges for mmu notifiers. Ensure that this range is flushed if huge_pmd_unshare succeeds and unmaps a PUD_SUZE area. Link: http://lkml.kernel.org/r/20180823205917.16297-3-mike.kravetz@xxxxxxxxxx Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx> Acked-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx> Reviewed-by: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> Cc: Vlastimil Babka <vbabka@xxxxxxx> Cc: Davidlohr Bueso <dave@xxxxxxxxxxxx> Cc: Michal Hocko <mhocko@xxxxxxxxxx> Cc: Jerome Glisse <jglisse@xxxxxxxxxx> Cc: Mike Kravetz <mike.kravetz@xxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- mm/hugetlb.c | 53 +++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 43 insertions(+), 10 deletions(-) --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -3393,8 +3393,8 @@ void __unmap_hugepage_range(struct mmu_g struct page *page; struct hstate *h = hstate_vma(vma); unsigned long sz = huge_page_size(h); - const unsigned long mmun_start = start; /* For mmu_notifiers */ - const unsigned long mmun_end = end; /* For mmu_notifiers */ + unsigned long mmun_start = start; /* For mmu_notifiers */ + unsigned long mmun_end = end; /* For mmu_notifiers */ bool force_flush = false; WARN_ON(!is_vm_hugetlb_page(vma)); @@ -3402,6 +3402,11 @@ void __unmap_hugepage_range(struct mmu_g BUG_ON(end & ~huge_page_mask(h)); tlb_start_vma(tlb, vma); + + /* + * If sharing possible, alert mmu notifiers of worst case. + */ + adjust_range_if_pmd_sharing_possible(vma, &mmun_start, &mmun_end); mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); address = start; for (; address < end; address += sz) { @@ -3512,12 +3517,23 @@ void unmap_hugepage_range(struct vm_area { struct mm_struct *mm; struct mmu_gather tlb; + unsigned long tlb_start = start; + unsigned long tlb_end = end; + + /* + * If shared PMDs were possibly used within this vma range, adjust + * start/end for worst case tlb flushing. + * Note that we can not be sure if PMDs are shared until we try to + * unmap pages. However, we want to make sure TLB flushing covers + * the largest possible range. + */ + adjust_range_if_pmd_sharing_possible(vma, &tlb_start, &tlb_end); mm = vma->vm_mm; - tlb_gather_mmu(&tlb, mm, start, end); + tlb_gather_mmu(&tlb, mm, tlb_start, tlb_end); __unmap_hugepage_range(&tlb, vma, start, end, ref_page); - tlb_finish_mmu(&tlb, start, end); + tlb_finish_mmu(&tlb, tlb_start, tlb_end); } /* @@ -4205,11 +4221,21 @@ unsigned long hugetlb_change_protection( pte_t pte; struct hstate *h = hstate_vma(vma); unsigned long pages = 0; + unsigned long f_start = start; + unsigned long f_end = end; + bool shared_pmd = false; + + /* + * In the case of shared PMDs, the area to flush could be beyond + * start/end. Set f_start/f_end to cover the maximum possible + * range if PMD sharing is possible. + */ + adjust_range_if_pmd_sharing_possible(vma, &f_start, &f_end); BUG_ON(address >= end); - flush_cache_range(vma, address, end); + flush_cache_range(vma, f_start, f_end); - mmu_notifier_invalidate_range_start(mm, start, end); + mmu_notifier_invalidate_range_start(mm, f_start, f_end); i_mmap_lock_write(vma->vm_file->f_mapping); for (; address < end; address += huge_page_size(h)) { spinlock_t *ptl; @@ -4220,6 +4246,7 @@ unsigned long hugetlb_change_protection( if (huge_pmd_unshare(mm, &address, ptep)) { pages++; spin_unlock(ptl); + shared_pmd = true; continue; } pte = huge_ptep_get(ptep); @@ -4254,12 +4281,18 @@ unsigned long hugetlb_change_protection( * Must flush TLB before releasing i_mmap_rwsem: x86's huge_pmd_unshare * may have cleared our pud entry and done put_page on the page table: * once we release i_mmap_rwsem, another task can do the final put_page - * and that page table be reused and filled with junk. + * and that page table be reused and filled with junk. If we actually + * did unshare a page of pmds, flush the range corresponding to the pud. */ - flush_hugetlb_tlb_range(vma, start, end); - mmu_notifier_invalidate_range(mm, start, end); + if (shared_pmd) { + flush_hugetlb_tlb_range(vma, f_start, f_end); + mmu_notifier_invalidate_range(mm, f_start, f_end); + } else { + flush_hugetlb_tlb_range(vma, start, end); + mmu_notifier_invalidate_range(mm, start, end); + } i_mmap_unlock_write(vma->vm_file->f_mapping); - mmu_notifier_invalidate_range_end(mm, start, end); + mmu_notifier_invalidate_range_end(mm, f_start, f_end); return pages << h->order; }
