Re: [PATCH v2 09/12] mm: hwpoison: soft offline supports thp migration

[Balbir Singh]

On 11/11/16 10:58, Naoya Horiguchi wrote:
> On Thu, Nov 10, 2016 at 09:31:10PM +1100, Balbir Singh wrote:
>>
>>
>> On 08/11/16 10:31, Naoya Horiguchi wrote:
>>> This patch enables thp migration for soft offline.
>>>
>>> Signed-off-by: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
>>> ---
>>>  mm/memory-failure.c | 31 ++++++++++++-------------------
>>>  1 file changed, 12 insertions(+), 19 deletions(-)
>>>
>>> diff --git v4.9-rc2-mmotm-2016-10-27-18-27/mm/memory-failure.c v4.9-rc2-mmotm-2016-10-27-18-27_patched/mm/memory-failure.c
>>> index 19e796d..6cc8157 100644
>>> --- v4.9-rc2-mmotm-2016-10-27-18-27/mm/memory-failure.c
>>> +++ v4.9-rc2-mmotm-2016-10-27-18-27_patched/mm/memory-failure.c
>>> @@ -1485,7 +1485,17 @@ static struct page *new_page(struct page *p, unsigned long private, int **x)
>>>  	if (PageHuge(p))
>>>  		return alloc_huge_page_node(page_hstate(compound_head(p)),
>>>  						   nid);
>>> -	else
>>> +	else if (thp_migration_supported() && PageTransHuge(p)) {
>>> +		struct page *thp;
>>> +
>>> +		thp = alloc_pages_node(nid,
>>> +			(GFP_TRANSHUGE | __GFP_THISNODE) & ~__GFP_RECLAIM,
>>> +			HPAGE_PMD_ORDER);
>>> +		if (!thp)
>>> +			return NULL;
>>
>> Just wondering if new_page() fails, migration of that entry fails. Do we then
>> split and migrate? I guess this applies to THP migration in general.
> 
> Yes, that's not implemented yet, but can be helpful.
> 
> I think that there are 2 types of callers of page migration,
> one is a caller that specifies the target pages individually (like move_pages
> and soft offline), and another is a caller that specifies the target pages
> by (physical/virtual) address range basis.
> Maybe the former ones want to fall back immediately to split and retry if
> thp migration fails, and the latter ones want to retry thp migration more.
> If this makes sense, we can make some more changes on retry logic to fit
> the situation.
> 

I think we definitely need the retry with split option, but may be we can
build it on top of this series.

Balbir Singh.