Re: [PATCHv6] blk-integrity: support arbitrary buffer alignment

[Keith Busch <kbusch@kernel.org>]

On Tue, Nov 25, 2025 at 12:31:44PM +0100, Christoph Hellwig wrote:
> On Mon, Nov 24, 2025 at 07:41:16PM -0800, Caleb Sander Mateos wrote:
> > > Compare kmap_local_page() with kmap_local_folio(). They both resolve to
> > > the same lower level mapping function, and folios have no problem
> > > spanning pages.
> > 
> > Documentation/mm/highmem.rst seems to suggest that kmap_local_folio()
> > only maps the page that contains the specified offset, not the whole
> > folio:
> 
> Yes.  Additionally multi-page bvecs aren't folios, they just are
> physically contiguous ranges.  So I don't think we could even use any
> folio helpers if they worked, as a single range might consist of multiple
> folios.
> 
> (Sorry for not spotting this earlier)

So the solution is to replace mp_bvec_iter_bvec() with bvec_iter_bvec()
to ensure we don't cross pages? It's a little less efficient, but that's
not a big deal.

I assumed mapping physically congiguous memory was contiguous in kernel
address space too, and that seems to work out in testing, but I don't
have CONFIG_HIGHMEM enable where that might matter.