Re: [PATCH v2 1/4] mm/gup: add compound page list iterator

[John Hubbard <jhubbard@nvidia.com>]

On 2/5/21 2:46 AM, Joao Martins wrote:
...>> If instead you keep npages constant like it naturally wants to be, you could
>> just do a "(*ntails)++" in the loop, to take care of *ntails.
>>
> I didn't do it as such as I would need to deref @ntails per iteration, so
> it felt more efficient to do as above. On a second thought, I could alternatively do the
> following below, thoughts?
> 
> diff --git a/mm/gup.c b/mm/gup.c
> index d68bcb482b11..8defe4f670d5 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -215,6 +215,32 @@ void unpin_user_page(struct page *page)
>   }
>   EXPORT_SYMBOL(unpin_user_page);
> 
> +static inline void compound_next(unsigned long i, unsigned long npages,
> +                                struct page **list, struct page **head,
> +                                unsigned int *ntails)
> +{
> +       struct page *page;
> +       unsigned int nr;
> +
> +       if (i >= npages)
> +               return;
> +
> +       page = compound_head(list[i]);
> +       for (nr = i + 1; nr < npages; nr++) {
> +               if (compound_head(list[nr]) != page)
> +                       break;
> +       }
> +
> +       *head = page;
> +       *ntails = nr - i;
> +}
> +

Yes, this is cleaner and quite a bit easier to verify that it is correct.

> 
>> However, given that the patch is correct and works as-is, the above is really just
>> an optional idea, so please feel free to add:
>>
>> Reviewed-by: John Hubbard <jhubbard@nvidia.com>
>>
>>
> Thanks!
> 
> Hopefully I can retain that if the snippet above is preferred?
> 
> 	Joao
> 

Yes. Still looks good.

thanks,
-- 
John Hubbard
NVIDIA