Eric Lee / smarc-ti-linux-kernel

Download as
Browse Code ยป

Commit 1ad9f620c3a22fa800489455ce517c29e576934e

Authored by Mel Gorman
Committed by Linus Torvalds
1 parent 88a9ab6e3d
Exists in master and in 13 other branches dlt-processor-sdk-linux-03.00.00.04, processor-sdk-linux-02.00.01, smarc-ti-linux-3.15.y, smarc-ti-lsk-linux-4.1.y, smarct3x-processor-sdk-04.01.00.06, smarct3x-processor-sdk-linux-02.00.01, smarct3x-processor-sdk-linux-03.00.00.04, smarct4x-800-processor-sdk-linux-02.00.01, smarct4x-processor-sdk-04.01.00.06, smarct4x-processor-sdk-linux-02.00.01, smarct4x-processor-sdk-linux-03.00.00.04, ti-linux-3.15.y, ti-lsk-linux-4.1.y

mm: numa: recheck for transhuge pages under lock during protection changes 

Sasha reported the following bug using trinity

  kernel BUG at mm/mprotect.c:149!
  invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
  Dumping ftrace buffer:
     (ftrace buffer empty)
  Modules linked in:
  CPU: 20 PID: 26219 Comm: trinity-c216 Tainted: G        W    3.14.0-rc5-next-20140305-sasha-00011-ge06f5f3-dirty #105
  task: ffff8800b6c80000 ti: ffff880228436000 task.ti: ffff880228436000
  RIP: change_protection_range+0x3b3/0x500
  Call Trace:
    change_protection+0x25/0x30
    change_prot_numa+0x1b/0x30
    task_numa_work+0x279/0x360
    task_work_run+0xae/0xf0
    do_notify_resume+0x8e/0xe0
    retint_signal+0x4d/0x92

The VM_BUG_ON was added in -mm by the patch "mm,numa: reorganize
change_pmd_range".  The race existed without the patch but was just
harder to hit.

The problem is that a transhuge check is made without holding the PTL.
It's possible at the time of the check that a parallel fault clears the
pmd and inserts a new one which then triggers the VM_BUG_ON check.  This
patch removes the VM_BUG_ON but fixes the race by rechecking transhuge
under the PTL when marking page tables for NUMA hinting and bailing if a
race occurred.  It is not a problem for calls to mprotect() as they hold
mmap_sem for write.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Showing 1 changed file with 34 additions and 2 deletions Side-by-side Diff

... ... @@ -36,6 +36,34 @@
36 36 }
37 37 #endif
38 38  
  39 +/*
  40 + * For a prot_numa update we only hold mmap_sem for read so there is a
  41 + * potential race with faulting where a pmd was temporarily none. This
  42 + * function checks for a transhuge pmd under the appropriate lock. It
  43 + * returns a pte if it was successfully locked or NULL if it raced with
  44 + * a transhuge insertion.
  45 + */
  46 +static pte_t *lock_pte_protection(struct vm_area_struct *vma, pmd_t *pmd,
  47 + unsigned long addr, int prot_numa, spinlock_t **ptl)
  48 +{
  49 + pte_t *pte;
  50 + spinlock_t *pmdl;
  51 +
  52 + /* !prot_numa is protected by mmap_sem held for write */
  53 + if (!prot_numa)
  54 + return pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
  55 +
  56 + pmdl = pmd_lock(vma->vm_mm, pmd);
  57 + if (unlikely(pmd_trans_huge(*pmd) || pmd_none(*pmd))) {
  58 + spin_unlock(pmdl);
  59 + return NULL;
  60 + }
  61 +
  62 + pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
  63 + spin_unlock(pmdl);
  64 + return pte;
  65 +}
  66 +
39 67 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
40 68 unsigned long addr, unsigned long end, pgprot_t newprot,
41 69 int dirty_accountable, int prot_numa)
... ... @@ -45,7 +73,10 @@
45 73 spinlock_t *ptl;
46 74 unsigned long pages = 0;
47 75  
48   - pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  76 + pte = lock_pte_protection(vma, pmd, addr, prot_numa, &ptl);
  77 + if (!pte)
  78 + return 0;
  79 +
49 80 arch_enter_lazy_mmu_mode();
50 81 do {
51 82 oldpte = *pte;
52 83  
... ... @@ -132,12 +163,13 @@
132 163 pages += HPAGE_PMD_NR;
133 164 nr_huge_updates++;
134 165 }
  166 +
  167 + /* huge pmd was handled */
135 168 continue;
136 169 }
137 170 }
138 171 /* fall through, the trans huge pmd just split */
139 172 }
140   - VM_BUG_ON(pmd_trans_huge(*pmd));
141 173 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
142 174 dirty_accountable, prot_numa);
143 175 pages += this_pages;