Re: [PATCH v2 2/3] mm/damon/paddr: minor refactor of damon_pa_young()

[SeongJae Park <sj@kernel.org>]

Hi Kefeng,

On Mon, 6 Mar 2023 09:56:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> 
> 
> On 2023/3/6 9:10, Kefeng Wang wrote:
> > 
> > 
> > On 2023/3/4 2:39, SeongJae Park wrote:
> >> Hi Kefeng,
> >>
> >> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang 
> >> <wangkefeng.wang@huawei.com> wrote:
> >>
> >>> Omit three lines by unified folio_put(), and make code more clear.
> >>>
> >>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >>> ---
> >>>   mm/damon/paddr.c | 11 ++++-------
> >>>   1 file changed, 4 insertions(+), 7 deletions(-)
> >>>
> >>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
> >>> index 3fda00a0f786..2ef9db0189ca 100644
> >>> --- a/mm/damon/paddr.c
> >>> +++ b/mm/damon/paddr.c
> >>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr, 
> >>> unsigned long *folio_sz)
> >>>               accessed = false;
> >>>           else
> >>>               accessed = true;
> >>> -        folio_put(folio);
> >>>           goto out;
> >>
> >> Because you moved 'out' label to not include *folio_sz setting, 
> >> folio_sz will
> >> not set in this case.  It should be set.
> > oh, it should be fixed.
> >>
> >>>       }
> >>>       need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
> >>> -    if (need_lock && !folio_trylock(folio)) {
> >>> -        folio_put(folio);
> >>> -        return false;
> >>> -    }
> 
> Hi SJ,  apart from above issue, it looks that this branch need the 
> folio_size() setting, right?

folio_sz is effectively used by caller of damon_pa_young() only if this
function returns true, so this branch doesn't need to set folio_sz.


Thanks,
SJ

> 
> Thanks
> 
> >>> +    if (need_lock && !folio_trylock(folio))
> >>> +        goto out;
> >>>       rmap_walk(folio, &rwc);
> >>>       if (need_lock)
> >>>           folio_unlock(folio);
> >>> -    folio_put(folio);
> >>> -out:
> >>>       *folio_sz = folio_size(folio);
> >>> +out:
> >>> +    folio_put(folio);
> >>
> >> Before this change, folio_size() is called after folio_put().  
> >> Shouldn't it be
> >> called before folio_put()?  If so, could we make a separate fix for 
> >> that first,
> >> and then make this change on top of it, so that it can be easily 
> >> applied to
> >> relevant stable kernels?
> >>
> > Yes， I could separate it, after folio_put(), the folio could be 
> > re-allocated and the folio_size calculation is not right.
> >>
> >> Thanks,
> >> SJ
> >>
> >>>       return accessed;
> >>>   }
> >>> -- 
> >>> 2.35.3
> >>>
> >>>
>