Re: [PATCH RFC 00/32] mm/mglru: MGLRU-FG and refault distance support

[Joanne Koong]

On Thu, Jul 23, 2026 at 1:38 AM Kairui Song <ryncsn@xxxxxxxxx> wrote:
>
> Joanne Koong <joannelkoong@xxxxxxxxx> 于 2026年7月23日周四 08:49写道：
> >
> > Gotcha. Can I ask what speaks against having the flags unification
> > work ordered first, with the MGLRU-FG patches based on top of that and
> > submitted in parallel as a separate series? It seems like the two
> > could be reviewed upstream concurrently, and I think that would make
> > the MGLRU-FG series easier to review with it on top of the flags
> > cleanup, and freeing the two page flags seems like a sweet win on its
> > own :) Is your concern that the FG series would then be gated on the
> > flag series landing and you anticipate that the flag series would take
> > longer to land than the FG series?
>
> Hi,
>
> That's not how it works, though. Without the MGLRU-FG, it takes four
> bits for tracking the tier and these two flag bits have special
> meanings. After FG, only three bits are needed and these two original
> flag bits are just the lower two bits of a three-bit counter now. So,
> there is an advantage to rearranging them together and removing the
> single-bit (flag) operations to always use them as a whole. Before
> that, that conversion will actual make thing more ugly.

Hi Kairui,

That makes sense, thanks for the context.

>
> BTW will there be any feature change with your splitting plan? If
> there is no feature change, that should be easy to review.

There will be no behavior change with the splitting. It will only be
structural.

Thanks,
Joanne