[PATCH v4 14/25] rmap: Add support for PUD sized mappings to rmap

[David Hildenbrand]

>>>    				return -EBUSY;
>>> diff --git a/mm/rmap.c b/mm/rmap.c
>>> index c6c4d4e..39d0439 100644
>>> --- a/mm/rmap.c
>>> +++ b/mm/rmap.c
>>> @@ -1203,6 +1203,11 @@ static __always_inline unsigned int __folio_add_rmap(struct folio *folio,
>>>    		}
>>>    		atomic_inc(&folio->_large_mapcount);
>>>    		break;
>>> +	case RMAP_LEVEL_PUD:
>>> +		/* We only support entire mappings of PUD sized folios in rmap */
>>> +		atomic_inc(&folio->_entire_mapcount);
>>> +		atomic_inc(&folio->_large_mapcount);
>>> +		break;
>>
>>
>> This way you don't account the pages at all as mapped, whereby PTE-mapping it
>> would? And IIRC, these PUD-sized pages can be either mapped using PTEs or
>> using a single PUD.
> 
> Oh good point. I was thinking that because we don't account PUD mappings today
> that it would be fine to ignore them. But of course this series means we start
> accounting them if mapped with PTEs so agree we should be consistent.
>   
>> I suspect what you want is to
> 
> Yes, I think so. Thanks for the hint. I will be out over the Christmas break but
> will do a respin to incorporate this before then.

I'll be on PTO starting ... well now. But I'll try to give the other 
parts a quick peek if anything urgent jumps at me. (bad habit of reading 
mails while on PTO ...)

In any case, happy holidays to you!

-- 
Cheers,

David / dhildenb