From: Nadav Amit namit@vmware.com
commit a4a118f2eead1d6c49e00765de89878288d4b890 upstream.
When __unmap_hugepage_range() calls to huge_pmd_unshare() succeed, a TLB flush is missing. This TLB flush must be performed before releasing the i_mmap_rwsem, in order to prevent an unshared PMDs page from being released and reused before the TLB flush took place.
Arguably, a comprehensive solution would use mmu_gather interface to batch the TLB flushes and the PMDs page release, however it is not an easy solution: (1) try_to_unmap_one() and try_to_migrate_one() also call huge_pmd_unshare() and they cannot use the mmu_gather interface; and (2) deferring the release of the page reference for the PMDs page until after i_mmap_rwsem is dropeed can confuse huge_pmd_unshare() into thinking PMDs are shared when they are not.
Fix __unmap_hugepage_range() by adding the missing TLB flush, and forcing a flush when unshare is successful.
Fixes: 24669e58477e ("hugetlb: use mmu_gather instead of a temporary linked list for accumulating pages)" # 3.6 Signed-off-by: Nadav Amit namit@vmware.com Reviewed-by: Mike Kravetz mike.kravetz@oracle.com Cc: Aneesh Kumar K.V aneesh.kumar@linux.vnet.ibm.com Cc: KAMEZAWA Hiroyuki kamezawa.hiroyu@jp.fujitsu.com Cc: Andrew Morton akpm@linux-foundation.org Signed-off-by: Linus Torvalds torvalds@linux-foundation.org Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
--- include/asm-generic/tlb.h | 6 ++++++ mm/hugetlb.c | 19 +++++++++++++++++++ 2 files changed, 25 insertions(+)
--- a/include/asm-generic/tlb.h +++ b/include/asm-generic/tlb.h @@ -196,6 +196,12 @@ static inline void tlb_remove_check_page #define tlb_end_vma __tlb_end_vma #endif
+static inline void tlb_flush_pmd_range(struct mmu_gather *tlb, + unsigned long address, unsigned long size) +{ + __tlb_adjust_range(tlb, address, size); +} + #ifndef __tlb_remove_tlb_entry #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0) #endif --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -3386,6 +3386,7 @@ void __unmap_hugepage_range(struct mmu_g 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 */ + bool force_flush = false;
WARN_ON(!is_vm_hugetlb_page(vma)); BUG_ON(start & ~huge_page_mask(h)); @@ -3407,6 +3408,8 @@ void __unmap_hugepage_range(struct mmu_g ptl = huge_pte_lock(h, mm, ptep); if (huge_pmd_unshare(mm, &address, ptep)) { spin_unlock(ptl); + tlb_flush_pmd_range(tlb, address & PUD_MASK, PUD_SIZE); + force_flush = true; continue; }
@@ -3463,6 +3466,22 @@ void __unmap_hugepage_range(struct mmu_g } mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); tlb_end_vma(tlb, vma); + + /* + * If we unshared PMDs, the TLB flush was not recorded in mmu_gather. We + * could defer the flush until now, since by holding i_mmap_rwsem we + * guaranteed that the last refernece would not be dropped. But we must + * do the flushing before we return, as otherwise i_mmap_rwsem will be + * dropped and the last reference to the shared PMDs page might be + * dropped as well. + * + * In theory we could defer the freeing of the PMD pages as well, but + * huge_pmd_unshare() relies on the exact page_count for the PMD page to + * detect sharing, so we cannot defer the release of the page either. + * Instead, do flush now. + */ + if (force_flush) + tlb_flush_mmu(tlb); }
void __unmap_hugepage_range_final(struct mmu_gather *tlb,