Re: [PATCH v10 3/4] mm, thp: introduce FOLL_SPLIT_PMD

[Oleg Nesterov]

On 07/31, Song Liu wrote:
>
> > On Jul 31, 2019, at 8:18 AM, Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> >
> > Now, I don't understand why do we need pmd_trans_unstable() after
> > split_huge_pmd(huge-zero-pmd), but whatever reason we have, why can't we
> > unify both cases?
> >
> > IOW, could you explain why the path below is wrong?
>
> I _think_ the following patch works (haven't fully tested yet). But I am not
> sure whether this is the best. By separating the two cases, we don't duplicate
> much code. And it is clear that the two cases are handled differently.
> Therefore, I would prefer to keep these separate for now.

I disagree. I think this separation makes the code less readable/understandable.
Exactly because it handles two cases differently and it is absolutely not clear
why.

But I can't argue, please forget.

Oleg.