Re: [PATCH] mm: Compute mTHP order efficiently

[Barry Song <baohua@kernel.org>]

On Tue, Sep 17, 2024 at 11:36 AM Lance Yang <ioworker0@gmail.com> wrote:
>
> On Mon, Sep 16, 2024 at 9:25 PM Matthew Wilcox <willy@infradead.org> wrote:
> >
> > On Fri, Sep 13, 2024 at 02:49:02PM +0530, Dev Jain wrote:
> > > We use pte_range_none() to determine whether contiguous PTEs are empty
> > > for an mTHP allocation. Instead of iterating the while loop for every
> > > order, use some information, which is the first set PTE found, from the
> > > previous iteration, to eliminate some cases. The key to understanding
> > > the correctness of the patch is that the ranges we want to examine
> > > form a strictly decreasing sequence of nested intervals.
> >
> > This is a lot more complicated.  Do you have any numbers that indicate
> > that it's faster?  Yes, it's fewer memory references, but you've gone
> > from a simple linear scan that's easy to prefetch to an exponential scan
> > that might confuse the prefetchers.
>
> +1
>
> I'm not sure if multiple mthp sizes will be enabled for common cases ;)
> If not, this could be a bit more complicated, IMO.
>
> @Barry, could you share whether OPPO typically uses multiple mthp sizes
> in their scenarios?

not at all. I actually doubt we really wanted to enable multiple sizes at
the same time.

but somehow, i like the idea to make pte_range_none() non-bool(to be
consistent with all other places we are checking pte/swap_xxx_batch()
!= nr_pages). Personally, I don't care about the overhead of doing PTE
scanning too much. so i'd like having patch1 converting to non-bool, for
patch2 which might/might not reduce overhead, i don't care too much.

>
> Thanks,
> Lance

Thanks
Barry