Re: [PATCH 10/12] xfs: use vmalloc instead of vm_map_area for buffer backing memory

[Dave Chinner <david@fromorbit.com>]

On Thu, Mar 06, 2025 at 12:35:36AM +0100, Christoph Hellwig wrote:
> On Thu, Mar 06, 2025 at 08:20:08AM +1100, Dave Chinner wrote:
> > > +		__bio_add_page(bio, virt_to_page(bp->b_addr),
> > > +				BBTOB(bp->b_length),
> > > +				offset_in_page(bp->b_addr));
> > >  	}
> > 
> > How does offset_in_page() work with a high order folio? It can only
> > return a value between 0 and (PAGE_SIZE - 1).
> 
> Yes.
> 
> > i.e. shouldn't this
> > be:
> > 
> > 		folio = kmem_to_folio(bp->b_addr);
> > 
> > 		bio_add_folio_nofail(bio, folio, BBTOB(bp->b_length),
> > 				offset_in_folio(folio, bp->b_addr));
> > 
> 
> That is also correct, but does a lot more work underneath as the
> bio_vecs work in terms of pages.  In the long run this should use
> a bio_add_virt that hides all that (and the bio_vecs should move to
> store physical addresses).  For now the above is the simplest and
> most efficient version.

Can you add a comment that the code is done this way because
of the mismatch between block layer API and mm object (folio/slab)
handling APIs? Otherwise this code is going to look wrong every time
I look at in future....

-Dave.
-- 
Dave Chinner
david@fromorbit.com