Commit 1ad9f620c3a22fa800489455ce517c29e576934e
Committed by
Linus Torvalds
1 parent
88a9ab6e3d
Exists in
master
and in
13 other branches
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
mm/mprotect.c
... | ... | @@ -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; |