From: David Gibson The page fault path for normal pages, if the fault is neither a no-page fault nor a write-protect fault, will update the DIRTY and ACCESSED bits in the page table appropriately. The hugepage fault path, however, does not do this, handling only no-page or write-protect type faults. It assumes that either the ACCESSED and DIRTY bits are irrelevant for hugepages (usually true, since they are never swapped) or that they are handled by the arch code. This is inconvenient for some software-loaded TLB architectures, where the _PAGE_ACCESSED (_PAGE_DIRTY) bits need to be set to enable read (write) access to the page at the TLB miss. This could be worked around in the arch TLB miss code, but the TLB miss fast path can be made simple more easily if the hugetlb_fault() path handles this, as the normal page fault path does. Signed-off-by: David Gibson Cc: William Lee Irwin III Cc: Hugh Dickins Cc: Adam Litke Signed-off-by: Andrew Morton --- mm/hugetlb.c | 23 ++++++++++++++++++----- 1 file changed, 18 insertions(+), 5 deletions(-) diff -puN mm/hugetlb.c~hugetlb-handle-updating-of-accessed-and-dirty-in-hugetlb_fault mm/hugetlb.c --- a/mm/hugetlb.c~hugetlb-handle-updating-of-accessed-and-dirty-in-hugetlb_fault +++ a/mm/hugetlb.c @@ -2008,7 +2008,7 @@ int hugetlb_fault(struct mm_struct *mm, entry = huge_ptep_get(ptep); if (huge_pte_none(entry)) { ret = hugetlb_no_page(mm, vma, address, ptep, write_access); - goto out_unlock; + goto out_mutex; } ret = 0; @@ -2024,7 +2024,7 @@ int hugetlb_fault(struct mm_struct *mm, if (write_access && !pte_write(entry)) { if (vma_needs_reservation(h, vma, address) < 0) { ret = VM_FAULT_OOM; - goto out_unlock; + goto out_mutex; } if (!(vma->vm_flags & VM_SHARED)) @@ -2034,10 +2034,23 @@ int hugetlb_fault(struct mm_struct *mm, spin_lock(&mm->page_table_lock); /* Check for a racing update before calling hugetlb_cow */ - if (likely(pte_same(entry, huge_ptep_get(ptep)))) - if (write_access && !pte_write(entry)) + if (unlikely(!pte_same(entry, huge_ptep_get(ptep)))) + goto out_page_table_lock; + + + if (write_access) { + if (!pte_write(entry)) { ret = hugetlb_cow(mm, vma, address, ptep, entry, pagecache_page); + goto out_page_table_lock; + } + entry = pte_mkdirty(entry); + } + entry = pte_mkyoung(entry); + if (huge_ptep_set_access_flags(vma, address, ptep, entry, write_access)) + update_mmu_cache(vma, address, entry); + +out_page_table_lock: spin_unlock(&mm->page_table_lock); if (pagecache_page) { @@ -2045,7 +2058,7 @@ int hugetlb_fault(struct mm_struct *mm, put_page(pagecache_page); } -out_unlock: +out_mutex: mutex_unlock(&hugetlb_instantiation_mutex); return ret; _