Re: [RFC PATCH 3/5] mm, hugetlb: do not rely on overcommit limit during migration

From: Mike Kravetz
Date: Thu Dec 14 2017 - 15:58:13 EST


On 12/13/2017 11:40 PM, Michal Hocko wrote:
> On Wed 13-12-17 15:35:33, Mike Kravetz wrote:
>> On 12/04/2017 06:01 AM, Michal Hocko wrote:
> [...]
>>> Before migration
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/free_hugepages:0
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages:1
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/surplus_hugepages:0
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/free_hugepages:0
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages:0
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/surplus_hugepages:0
>>>
>>> After
>>>
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/free_hugepages:0
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages:0
>>> /sys/devices/system/node/node0/hugepages/hugepages-2048kB/surplus_hugepages:0
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/free_hugepages:0
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages:1
>>> /sys/devices/system/node/node1/hugepages/hugepages-2048kB/surplus_hugepages:0
>>>
>>> with the previous implementation, both nodes would have nr_hugepages:1
>>> until the page is freed.
>>
>> With the previous implementation, the migration would have failed unless
>> nr_overcommit_hugepages was explicitly set. Correct?
>
> yes
>
> [...]
>
>> In the previous version of this patch, I asked about handling of 'free' huge
>> pages. I did a little digging and IIUC, we do not attempt migration of
>> free huge pages. The routine isolate_huge_page() has this check:
>>
>> if (!page_huge_active(page) || !get_page_unless_zero(page)) {
>> ret = false;
>> goto unlock;
>> }
>>
>> I believe one of your motivations for this effort was memory offlining.
>> So, this implies that a memory area can not be offlined if it contains
>> a free (not in use) huge page?
>
> do_migrate_range will ignore this free huge page and then we will free
> it up in dissolve_free_huge_pages
>
>> Just FYI and may be something we want to address later.
>
> Maybe yes. The free pool might be reserved which would make
> dissolve_free_huge_pages to fail. Maybe we can be more clever and
> allocate a new huge page in that case.

Don't think we need to try and do anything more clever right now. I was
just a little confused about the hot plug code. Thanks for the explanation.

--
Mike Kravetz

>
>> My other issues were addressed.
>>
>> Reviewed-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
>
> Thanks!
>