Re: swiotlb/caamjr regression (Was: [GIT PULL] (swiotlb) stable/for-linus-5.12)

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, Jun 11, 2021 at 3:35 AM Konrad Rzeszutek Wilk
<konrad.wilk@oracle.com> wrote:
>
> Linus,
>
> Would you be terribly offended if I took your code (s/unsigned
> long/unsigned int), and used Chanho's description of the problem (see below)?

No offense to that at all - that looks like the right solution. See my
answer to Christoph: I do think my patch does the right one, but I
can't test it and my knowledge of the swiotlb code is not complete
enough to really do anything else than "this looks right".

And adding my sign-off to the patch is fine, but I don't necessarily
need the authorship credit - mine was a throw-away patch just looking
at what the bisection report said. All the real effort was by the
reporters (and for the commit message, Bumyong Lee & co).

Finally - looking at the two places that do have that
swiotlb_align_offset(), my reaction is "I don't understand what that
code is doing".

The index does that

        index = find_slots(dev, orig_addr, alloc_size + offset);

so that offset does seem to depend on how the find_slots code works.
Which in turn does use the same dma_get_min_align_mask() thing that
swiotlb_align_offset() uses.  So the offsets do seem to match, but
find_slots(dev() does a lot of stuff that I don't know. so...

IOW, it does reinforce my "I don't know this code AT ALL". Which just
makes me more convinced that I shouldn't get authorship of the patch
because if something goes wrong with it, I can't help.

So at most maybe a "Suggested-by:".

My patch really was based on very little context and "this is the
calculation that makes sense given the other calculations in the
function".

              Linus