Re: [PATCH 1/1] mm: Add folio_map_local()

[David Howells <dhowells@redhat.com>]

Matthew Wilcox (Oracle) <willy@infradead.org> wrote:

> Some filesystems benefit from being able to map the entire folio.
> On 32-bit platforms with HIGHMEM, we fall back to using vmap, which
> will be slow.  If it proves to be a performance problem, we can look at
> optimising it in a number of ways.

Here's a thought: If a highmem arch has a huge PTEs available, can you create
a huge PTE to cover the folio?

> @@ -426,5 +465,4 @@ static inline void folio_zero_range(struct folio *folio,
>  {
>  	zero_user_segments(&folio->page, start, start + length, 0, 0);
>  }
> -
>  #endif /* _LINUX_HIGHMEM_H */

Did you want to remove that blank line?

David