Re: [PATCH v3 2/5] hugetlb: convert page_huge_active() HPageMigratable flag

[Mike Kravetz <mike.kravetz@oracle.com>]

On 1/27/21 2:25 AM, Michal Hocko wrote:
> On Fri 22-01-21 11:52:28, Mike Kravetz wrote:
>> Use the new hugetlb page specific flag HPageMigratable to replace the
>> page_huge_active interfaces.  By it's name, page_huge_active implied
>> that a huge page was on the active list.  However, that is not really
>> what code checking the flag wanted to know.  It really wanted to determine
>> if the huge page could be migrated.  This happens when the page is actually
>> added to the page cache and/or task page table.  This is the reasoning
>> behind the name change.
> 
> yeah, definitely less confusing!
> 
>> The VM_BUG_ON_PAGE() calls in the *_huge_active() interfaces are not
>> really necessary as we KNOW the page is a hugetlb page.  Therefore, they
>> are removed.
>>
>> The routine page_huge_active checked for PageHeadHuge before testing the
>> active bit.  This is unnecessary in the case where we hold a reference or
>> lock and know it is a hugetlb head page.  page_huge_active is also called
>> without holding a reference or lock (scan_movable_pages), and can race with
>> code freeing the page.  The extra check in page_huge_active shortened the
>> race window, but did not prevent the race.  Offline code calling
>> scan_movable_pages already deals with these races, so removing the check
>> is acceptable.  Add comment to racy code.
>>
>> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
>> Reviewed-by: Oscar Salvador <osalvador@suse.de>
>> Reviewed-by: Muchun Song <songmuchun@bytedance.com>
>> Reviewed-by: Miaohe Lin <linmiaohe@huawei.com>
> 
> Acked-by: Michal Hocko <mhocko@suse.com>
> 
> One nit below
> [...]
>> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
>> index a7eb05315c6e..58be44a915d1 100644
>> --- a/include/linux/hugetlb.h
>> +++ b/include/linux/hugetlb.h
>> @@ -480,9 +480,13 @@ unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
>>   * HPG_restore_reserve - Set when a hugetlb page consumes a reservation at
>>   *	allocation time.  Cleared when page is fully instantiated.  Free
>>   *	routine checks flag to restore a reservation on error paths.
>> + * HPG_migratable  - Set after a newly allocated page is added to the page
>> + *	cache and/or page tables.  Indicates the page is a candidate for
>> + *	migration.
> 
> The state change is synchronized by hugetlb_lock.
> 
> Pls.

I will update/explain this.  But ... hugetlb_lock does not synchronize all
changes of this flag.  The flag is set without holding the lock for newly
allocated pages after being added to page cache and/or page tables.  This
'signals' the page is a candidate for migration.  When the migration code
(isolation, putback, move state) deals with the flag, it does hold the
hugetlb_lock.

This patch did not change any of the synchronization.
-- 
Mike Kravetz