Re: flush_dcache_page vs kunmap_local

On Thu, Nov 4, 2021 at 11:39 AM Matthew Wilcox [off-list ref] wrote:

On Thu, Nov 04, 2021 at 08:30:55AM -0700, Linus Torvalds wrote:

quoted

Why did this come up? Do you actually have some hardware or situation
that cares?

Oh, we're doing review of the XFS/iomap folio patches, which led to
looking at zero_user_segments(), and I realised that memzero_page()
was now functionally identical to zero_user().  And you'd been quite
specific about not having flush_dcache_page() in there, so ... I wondered
if you'd had a change of mind.

Ugh. I guess it ends up being there whether I like it or not. All that
"zero_user_segments() stuff is too ugly for words, though, so I think
whoever wrote it must have been on some interesting pharmaceuticals.

What the hell are the two start/end things? And most users actually
just want a single page and should never have used that thing. Nasty.

I'm not touching that with a ten-foot pole.

              Linus