]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
hugetlb: fix nr_pmds accounting with shared page tables
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Fri, 24 Jun 2016 21:49:51 +0000 (14:49 -0700)
committerJack Vogel <jack.vogel@oracle.com>
Wed, 31 Jan 2018 19:39:29 +0000 (11:39 -0800)
We account HugeTLB's shared page table to all processes who share it.
The accounting happens during huge_pmd_share().

If somebody populates pud entry under us, we should decrease pagetable's
refcount and decrease nr_pmds of the process.

By mistake, I increase nr_pmds again in this case.  :-/ It will lead to
"BUG: non-zero nr_pmds on freeing mm: 2" on process' exit.

Let's fix this by increasing nr_pmds only when we're sure that the page
table will be used.

Link: http://lkml.kernel.org/r/20160617122506.GC6534@node.shutemov.name
Fixes: dc6c9a35b66b ("mm: account pmd page tables to the process")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: zhongjiang <zhongjiang@huawei.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Orabug: 27451809

(cherry picked from commit c17b1f42594eb71b8d3eb5a6dfc907a7eb88a51d)
Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
mm/hugetlb.c

index 53f681443fa4ce16f01fab9d8a244765ca1f3341..d42fa7fcca8e5117f40833246b61faffb032f3f5 100644 (file)
@@ -4354,7 +4354,6 @@ pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud)
                if (saddr) {
                        spte = huge_pte_offset(svma->vm_mm, saddr);
                        if (spte) {
-                               mm_inc_nr_pmds(mm);
                                get_page(virt_to_page(spte));
                                break;
                        }
@@ -4369,9 +4368,9 @@ pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud)
        if (pud_none(*pud)) {
                pud_populate(mm, pud,
                                (pmd_t *)((unsigned long)spte & PAGE_MASK));
+               mm_inc_nr_pmds(mm);
        } else {
                put_page(virt_to_page(spte));
-               mm_inc_nr_pmds(mm);
        }
        spin_unlock(ptl);
 out: