Re: [RFC PATCH] mm: support large folio numa balancing

[Baolin Wang]

On 11/15/2023 6:47 PM, David Hildenbrand wrote:
> On 15.11.23 11:46, David Hildenbrand wrote:
> > On 13.11.23 11:45, Baolin Wang wrote:
> > > Currently, the file pages already support large folio, and supporting 
> > > for
> > > anonymous pages is also under discussion[1]. Moreover, the numa 
> > > balancing
> > > code are converted to use a folio by previous thread[2], and the 
> > > migrate_pages
> > > function also already supports the large folio migration.
> > >
> > > So now I did not see any reason to continue restricting NUMA 
> > > balancing for
> > > large folio.
> > >
> > > [1] https://lkml.org/lkml/2023/9/29/342
> > > [2] 
> > > https://lore.kernel.org/all/20230921074417.24004-4-wangkefeng.wang@xxxxxxxxxx/T/#md9d10fe34587229a72801f0d731f7457ab3f4a6e
> > > Signed-off-by: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>
> > > ---
> >
> > I'll note that another piece is missing, and I'd be curious how you
> > tested your patch set or what I am missing. (no anonymous pages?)

I tested it with file large folio (order = 4) created by XFS filesystem.

> > change_pte_range() contains:
> >
> > if (prot_numa) {
> >     ...
> >     /* Also skip shared copy-on-write pages */
> >     if (is_cow_mapping(vma->vm_flags) &&
> >         folio_ref_count(folio) != 1)
> >         continue;
> >
> > So we'll never end up mapping an anon PTE-mapped THP prot-none (well, 
> > unless a
> > single PTE remains) and consequently never trigger NUMA hinting faults.
> >
> > Now, that change has some history [1], but the original problem has been
> > sorted out in the meantime. But we should consider Linus' original 
> > feedback.
> >
> > For pte-mapped THP, we might want to do something like the following
> > (completely untested):

Thanks for pointing out. I have not tried pte-mapped THP yet, and will 
look at it in detail.

> > diff --git a/mm/mprotect.c b/mm/mprotect.c
> > index 81991102f785..c4e6b9032e40 100644
> > --- a/mm/mprotect.c
> > +++ b/mm/mprotect.c
> > @@ -129,7 +129,8 @@ static long change_pte_range(struct mmu_gather *tlb,
> >                                   /* Also skip shared copy-on-write 
> > pages */
> >                                   if (is_cow_mapping(vma->vm_flags) &&
> > -                                   folio_ref_count(folio) != 1)
> > +                                   (folio_maybe_dma_pinned(folio) ||
> > +                                    folio_estimated_sharers(folio) != 
> > 1))
>
> Actually, > 1 might be better if the first subpage is not mapped; it's a 
> mess.