Re: [PATCH V1 1/9] mm/gup: repin_folio_unhugely

[David Hildenbrand <david@redhat.com>]

On 26.09.24 13:40, Jason Gunthorpe wrote:
> On Thu, Sep 26, 2024 at 01:32:09PM +0200, David Hildenbrand wrote:
>> On 20.09.24 15:28, Jason Gunthorpe wrote:
>>> On Thu, Sep 19, 2024 at 10:11:38AM +0200, David Hildenbrand wrote:
>>>
>>>>>> And repin is really misleading. We are simply adding more pins to an already pinned one ...
>>>>>
>>>>> Jason suggests a better name in the other thread.
>>>>
>>>> I would prefer something that simply adds more pins to an already pinned
>>>> folio. Much easier to get.
>>>
>>> Yes, but also nobody should ever want to do that operation, it should
>>> always be part of some kind of "splitting" sort of behavior..
>>
>> I remember patches from Dave Howells that needed that for O_DIRECT handling.
>> Never say never ;)
> 
> Wouldn't O_DIRECT be the same splitting thing?

No, I recall the implementation had to duplicate pins, not split.

> 
>> Adding is much more intuitive than splitting ... just like we add references
>> when splitting a THP, using folio_ref_add().
> 
> Well, sure, it just seems harder to document so people can use it
> properly.

As soon as we have large GUP and splitting might go to multiple PMDs, 
multiple PTEs or a mixture, just being able to add the number of pins 
you actually need might be cleaner ...

Let me have a look at the latest reincarnation of this patch.

-- 
Cheers,

David / dhildenb