[PATCH] mm: split vm_normal_pages for LRU and non-LRU handling

[Felix Kuehling]

Am 2022-03-01 um 11:22 schrieb David Hildenbrand:
>>>>    		if (PageReserved(page))
>>>> diff --git a/mm/migrate.c b/mm/migrate.c
>>>> index c31d04b46a5e..17d049311b78 100644
>>>> --- a/mm/migrate.c
>>>> +++ b/mm/migrate.c
>>>> @@ -1614,7 +1614,7 @@ static int add_page_for_migration(struct mm_struct *mm, unsigned long addr,
>>>>    		goto out;
>>>>    
>>>>    	/* FOLL_DUMP to ignore special (like zero) pages */
>>>> -	follflags = FOLL_GET | FOLL_DUMP;
>>>> +	follflags = FOLL_GET | FOLL_DUMP | FOLL_LRU;
>>>>    	page = follow_page(vma, addr, follflags);
>>> Why wouldn't we want to dump DEVICE_COHERENT pages? This looks wrong.
>> This function later calls isolate_lru_page, which is something you can't
>> do with a device page.
>>
> Then, that code might require care instead. We most certainly don't want
> to have random memory holes in a dump just because some anonymous memory
> was migrated to DEVICE_COHERENT.
I don't think this code is for core dumps. The call chain I see is

SYSCALL_DEFINE6(move_pages, ...) -> kernel_move_pages -> do_pages_move 
-> add_page_for_migration

As far as I can tell this is for NUMA migrations.

Regards,
   Felix