Re: gigantic hugepages vs. movable zones

[Vlastimil Babka]

On 07/27/2017 10:22 AM, Michal Hocko wrote:
> [CC for real]
> 
> On Thu 27-07-17 10:12:36, Michal Hocko wrote:
>> On Thu 27-07-17 13:30:31, Aneesh Kumar K.V wrote:
>>>
>>>
>>> On 07/27/2017 12:58 PM, Michal Hocko wrote:
>>>> On Thu 27-07-17 07:52:08, Aneesh Kumar K.V wrote:
>>>>> Michal Hocko <mhocko@xxxxxxxxxx> writes:
>>>>>
>>>>>> Hi,
>>>>>> I've just noticed that alloc_gigantic_page ignores movability of the
>>>>>> gigantic page and it uses any existing zone. Considering that
>>>>>> hugepage_migration_supported only supports 2MB and pgd level hugepages
>>>>>> then 1GB pages are not migratable and as such allocating them from a
>>>>>> movable zone will break the basic expectation of this zone. Standard
>>>>>> hugetlb allocations try to avoid that by using htlb_alloc_mask and I
>>>>>> believe we should do the same for gigantic pages as well.
>>>>>>
>>>>>> I suspect this behavior is not intentional. What do you think about the
>>>>>> following untested patch?
>>>>>
>>>>>
>>>>> I also noticed an unrelated issue with the usage of
>>>>> start_isolate_page_range. On error we set the migrate type to
>>>>> MIGRATE_MOVABLE.
>>>>
>>>> Why that should be a problem? I think it is perfectly OK to have
>>>> MIGRATE_MOVABLE pageblocks inside kernel zones.
>>>>
>>>
>>> we can pick pages with migrate type movable and if we fail to isolate won't

                                        ^ CMA

>>> we set the migrate type of that pages to MOVABLE ?

Yes, it seems we can silently kill CMA pageblocks in such case. Joonsoo,
can you check?

>>
>> I do not see an immediate problem. GFP_KERNEL allocations can fallback
>> to movable migrate pageblocks AFAIR. But I am not very much familiar
>> with migratetypes. Vlastimil, could you have a look please?
>