Re: mm: hangs in collapse_huge_page

[Hillf Danton]

hi all

On Wed, Apr 30, 2014 at 11:42 PM, Kirill A. Shutemov
<kirill@xxxxxxxxxxxxx> wrote:
> On Tue, Apr 15, 2014 at 10:06:56PM -0400, Sasha Levin wrote:
>> Hi all,
>>
>> I often see hung task triggering in khugepaged within collapse_huge_page().
>>
>> I've initially assumed the case may be that the guests are too loaded and
>> the warning occurs because of load, but after increasing the timeout to
>> 1200 sec I still see the warning.
>
> I suspect it's race (although I didn't track down exact scenario) with
> __khugepaged_exit().
>
> Comment in __khugepaged_exit() says that khugepaged_test_exit() always
> called under mmap_sem:
>
> 2045 void __khugepaged_exit(struct mm_struct *mm)
> ...
> 2063         } else if (mm_slot) {
> 2064                 /*
> 2065                  * This is required to serialize against
> 2066                  * khugepaged_test_exit() (which is guaranteed to run
> 2067                  * under mmap sem read mode). Stop here (after we
> 2068                  * return all pagetables will be destroyed) until
> 2069                  * khugepaged has finished working on the pagetables
> 2070                  * under the mmap_sem.
> 2071                  */
> 2072                 down_write(&mm->mmap_sem);
> 2073                 up_write(&mm->mmap_sem);
> 2074         }
> 2075 }
>
> But this is not true. At least khugepaged_scan_mm_slot() calls it without
> the sem:
>
> 2566 static unsigned int khugepaged_scan_mm_slot(unsigned int pages,
> 2567                                             struct page **hpage)
> ...
> 2046 {
> 2047         struct mm_slot *mm_slot;
> 2048         int free = 0;
> 2049
> 2050         spin_lock(&khugepaged_mm_lock);
> 2051         mm_slot = get_mm_slot(mm);
> 2052         if (mm_slot && khugepaged_scan.mm_slot != mm_slot) {
> 2053                 hash_del(&mm_slot->hash);
> 2054                 list_del(&mm_slot->mm_node);
> 2055                 free = 1;
> 2056         }
> 2057         spin_unlock(&khugepaged_mm_lock);
> 2058
> 2059         if (free) {
> 2060                 clear_bit(MMF_VM_HUGEPAGE, &mm->flags);
> 2061                 free_mm_slot(mm_slot);
> 2062                 mmdrop(mm);
>
> Not sure yet if it's a real problem or not. Andrea, could you comment on
> this?
>
> Sasha, please try patch below.
>
This box is quite,

 CPU:  0 PID: 0 Comm: swapper/0
 CPU:  1 PID: 0 Comm: swapper/1
 CPU:  2 PID: 0 Comm: swapper/2
 CPU:  3 PID: 0 Comm: swapper/3
 CPU:  4 PID: 0 Comm: swapper/4
 CPU:  5 PID: 0 Comm: swapper/5
 CPU:  6 PID: 0 Comm: swapper/6
 CPU:  7 PID: 0 Comm: swapper/7
 CPU:  8 PID: 0 Comm: swapper/8
 CPU:  9 PID: 0 Comm: swapper/9
 CPU: 10 PID: 0 Comm: swapper/10
 CPU: 11 PID: 0 Comm: swapper/11
 CPU: 12 PID: 0 Comm: swapper/12
 CPU: 13 PID: 0 Comm: swapper/13
 CPU: 14 PID: 0 Comm: swapper/14
 CPU: 15 PID: 0 Comm: swapper/15
 CPU: 16 PID: 0 Comm: swapper/16
 CPU: 17 PID: 0 Comm: swapper/17
 CPU: 18 PID: 0 Comm: swapper/18
 CPU: 19 PID: 0 Comm: swapper/19
 CPU: 20 PID: 0 Comm: swapper/20
 CPU: 21 PID: 3540 Comm: khungtaskd
 CPU: 22 PID: 0 Comm: swapper/22
 CPU: 23 PID: 0 Comm: swapper/23

and lets make more noise.

Hillf
---

--- a/mm/huge_memory.c Thu May  1 22:20:20 2014
+++ b/mm/huge_memory.c Thu May  1 22:24:06 2014
@@ -2732,7 +2732,8 @@ static void khugepaged_wait_work(void)
  }

  if (khugepaged_enabled())
- wait_event_freezable(khugepaged_wait, khugepaged_wait_event());
+ wait_event_freezable_timeout(khugepaged_wait, khugepaged_wait_event(),
+ msecs_to_jiffies(2000));
 }

 static int khugepaged(void *none)
--
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/