Re: [PATCH 0/7] RFC: high-order folio support for I/O

[Matthew Wilcox <willy@infradead.org>]

On Wed, Jun 14, 2023 at 05:35:02PM +0200, Hannes Reinecke wrote:
> Hmm. And for that I'm hitting include/linux/pagemap.h:1250 pretty
> consistently; something's going haywire with readahead.

Is that this one?

        VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);

in __readahead_folio()

> Matthew, are you sure that this one:
> 
> /**
> 
>  * readahead_length - The number of bytes in this readahead request.
> 
>  * @rac: The readahead request.
> 
>  */
> static inline size_t readahead_length(struct readahead_control *rac)
> {
>         return rac->_nr_pages * PAGE_SIZE;
> }
> 
> is tenable for large folios?
> Especially as we have in mm/readahead.c:499
> 
>         ractl->_nr_pages += 1UL << order;
> 
> Hmm?

Yes.  nr_pages really is the number of pages.

I'm somewhat surprised you're not hitting:

                VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);

in __filemap_add_folio().  Maybe the filesystems in question always
access indices which are naturally aligned to the block size.