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

[Kefeng Wang]

On 2023/11/14 19:35, David Hildenbrand wrote:
> On 13.11.23 23:15, John Hubbard wrote:
> > On 11/13/23 5:01 AM, Baolin Wang wrote:
> > > On 11/13/2023 8:10 PM, Kefeng Wang wrote:
> > > > On 2023/11/13 18:53, 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.
> > > > >
> > > > > I recall John wanted to look into that. CCing him.
> > > > >
> > > > > I'll note that the "head page mapcount" heuristic to detect sharers 
> > > > > will
> > > > > now strike on the PTE path and make us believe that a large folios is
> > > > > exclusive, although it isn't.
> > > > >
> > > > > As spelled out in the commit you are referencing:
> > > > >
> > > > > commit 6695cf68b15c215d33b8add64c33e01e3cbe236c
> > > > > Author: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
> > > > > Date:   Thu Sep 21 15:44:14 2023 +0800
> > > > >
> > > > >       mm: memory: use a folio in do_numa_page()
> > > > >       Numa balancing only try to migrate non-compound page in
> > > > > do_numa_page(),
> > > > >       use a folio in it to save several compound_head calls, note 
> > > > > we use
> > > > >       folio_estimated_sharers(), it is enough to check the folio
> > > > > sharers since
> > > > >       only normal page is handled, if large folio numa balancing is
> > > > > supported, a
> > > > >       precise folio sharers check would be used, no functional change
> > > > > intended.
> > > > >
> > > > >
> > > > > I'll send WIP patches for one approach that can improve the situation
> > > > > soonish.
> >
> > To be honest, I'm still catching up on the approximate vs. exact
> > sharers case. It wasn't clear to me why a precise sharers count
> > is needed in order to do this. Perhaps the cost of making a wrong
> > decision is considered just too high?
>
> Good question, I didn't really look into the impact for the NUMA hinting 
> case where we might end up not setting TNF_SHARED although it is shared. 
> For other folio_estimate_sharers() users it's more obvious.

The task_numa_group() will check the TNF_SHARED, if processes share same
page/folio, they will be packed into a single numa group, and the numa
group fault statistic will be used in should_numa_migrate_memory() to
decide whether to migrate or not, if not setting TNF_SHARED, maybe be
lead to more page/folio migration.

> As a side note, it could have happened already in corner cases (e.g., 
> concurrent page migration of a small folio).
>
> If precision as documented in that commit is really required remains to 
> be seen -- just wanted to spell it out.