Re: [PATCH] mm/hugetlbfs: Unmap pages if page fault raced with hole punch

[Mike Kravetz]

On 01/11/2016 06:20 PM, Andrew Morton wrote:
> On Mon, 11 Jan 2016 17:36:43 -0800 Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
> 
>>>
>>> I'll mark this patch as "pending, awaiting Mike's go-ahead".
>>>
>>
>> When this patch was originally submitted, bugs were discovered in the
>> hugetlb_vmdelete_list routine.  So, the patch "Fix bugs in
>> hugetlb_vmtruncate_list" was created.
>>
>> I have retested the changes in this patch specifically dealing with
>> page fault/hole punch race on top of the new hugetlb_vmtruncate_list
>> routine.  Everything looks good.
>>
>> How would you like to proceed with the patch?
>> - Should I create a series with the hugetlb_vmtruncate_list split out?
>> - Should I respin with hugetlb_vmtruncate_list patch applied?
>>
>> Just let me know what is easiest/best for you.
> 
> If you're saying that
> http://ozlabs.org/~akpm/mmots/broken-out/mm-mempolicy-skip-non-migratable-vmas-when-setting-mpol_mf_lazy.patch

That should be,
http://ozlabs.org/~akpm/mmots/broken-out/mm-hugetlbfs-fix-bugs-in-hugetlb_vmtruncate_list.patch

> and
> http://ozlabs.org/~akpm/mmots/broken-out/mm-hugetlbfs-unmap-pages-if-page-fault-raced-with-hole-punch.patch
> are the final everything-works versions then we're all good to go now.
> 

The only thing that 'might' be an issue is the new reference to
hugetlb_vmdelete_list() from remove_inode_hugepages().
hugetlb_vmdelete_list() was after remove_inode_hugepages() in the source
file.

The original patch moved hugetlb_vmdelete_list() to satisfy the new
reference.  I can not tell if that was taken into account in the way the
patches were pulled into your tree.  Will certainly know when it comes
time to build.

-- 
Mike Kravetz