Re: [RFC 2/2] migrate: convert migrate_pages() to use folios

[Huang, Ying]

Matthew Wilcox <willy@xxxxxxxxxxxxx> writes:

> On Tue, Nov 01, 2022 at 02:21:37PM +0800, Huang Ying wrote:
>> -			is_thp = PageTransHuge(page) && !PageHuge(page);
>> -			nr_subpages = compound_nr(page);
>> +			is_large = folio_test_large(folio) && !PageHuge(&folio->page);
>
> We have folio_test_hugetlb() to replace PageHuge().

Thanks!  Will change this in the next version.

>>  	count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
>>  	count_vm_events(PGMIGRATE_FAIL, nr_failed_pages);
>> -	count_vm_events(THP_MIGRATION_SUCCESS, nr_thp_succeeded);
>> -	count_vm_events(THP_MIGRATION_FAIL, nr_thp_failed);
>> -	count_vm_events(THP_MIGRATION_SPLIT, nr_thp_split);
>> -	trace_mm_migrate_pages(nr_succeeded, nr_failed_pages, nr_thp_succeeded,
>> -			       nr_thp_failed, nr_thp_split, mode, reason);
>> +	count_vm_events(THP_MIGRATION_SUCCESS, nr_large_succeeded);
>> +	count_vm_events(THP_MIGRATION_FAIL, nr_large_failed);
>> +	count_vm_events(THP_MIGRATION_SPLIT, nr_split);
>> +	trace_mm_migrate_pages(nr_succeeded, nr_failed_pages, nr_large_succeeded,
>> +			       nr_large_failed, nr_split, mode, reason);
>
> I think this is the biggest question with this patch -- how (or whether)
> to account folios in size between PMD and PTE size.  Since it's
> _called_ THP, I've tended to make the statistics conditional on
> folio_test_pmd_mappable() rather than simply being folio_test_large().

Yes.  This is my biggest concern during development too.  In the next
version, I will distinguish THP and other large folios (between PMD and
PTE size) and do statistics accordingly.

Best Regards,
Huang, Ying