Re: [PATCH 1/6] Revert "mm: mempolicy: Let vma_merge and vma_splithandle vma->vm_policy linkages"

[KOSAKI Motohiro]

On Wed, May 30, 2012 at 3:17 PM, Christoph Lameter <cl@xxxxxxxxx> wrote:
> On Wed, 30 May 2012, kosaki.motohiro@xxxxxxxxx wrote:
>
>> From: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxx>
>>
>> commit 05f144a0d5 removed vma->vm_policy updates code and it is a purpose of
>> mbind_range(). Now, mbind_range() is virtually no-op. no-op function don't
>> makes any bugs, I agree. but maybe it is not right fix.
>
> I dont really understand the changelog. But to restore the policy_vma() is
> the right thing to do since there are potential multiple use cases where
> we want to apply a policy to a vma.
>
> Proposed new changelog:
>
> Commit 05f144a0d5 folded policy_vma() into mbind_range(). There are
> other use cases of policy_vma(*) though and so revert a piece of
> that commit in order to have a policy_vma() function again.
>
>> @@ -655,23 +676,9 @@ static int mbind_range(struct mm_struct *mm, unsigned long start,
>>                       if (err)
>>                               goto out;
>>               }
>> -
>> -             /*
>> -              * Apply policy to a single VMA. The reference counting of
>> -              * policy for vma_policy linkages has already been handled by
>> -              * vma_merge and split_vma as necessary. If this is a shared
>> -              * policy then ->set_policy will increment the reference count
>> -              * for an sp node.
>> -              */
>
> You are dropping the nice comments by Mel that explain the refcounting.

Because this is not strictly correct. 1) vma_merge() and split_vma() don't
care mempolicy refcount. They only dup and drop it. 2) This mpol_get() is
for vma attaching. This function don't need to care sp_node internal.
--
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/