RE: [PATCH v2] mm/slub: Simplify get_partial_node()

[David Rientjes]

On Wed, 13 Mar 2024, Song, Xiongwei wrote:

> > On 2024/3/12 22:05, sxwjean@xxxxxx wrote:
> > > From: Xiongwei Song <xiongwei.song@xxxxxxxxxxxxx>
> > >
> > > Remove the check of !kmem_cache_has_cpu_partial() because it is always
> > > false, we've known this by calling kmem_cache_debug() before calling
> > > remove_partial(), so we can remove the check.
> > >
> > > Meanwhile, redo filling cpu partial and add comment to improve the
> > > readability.
> > >
> > > Signed-off-by: Xiongwei Song <xiongwei.song@xxxxxxxxxxxxx>
> > > ---
> > > Changes in v2:
> > >  - Use "#if IS_ENABLED(CONFIG_SLUB_CPU_PARTIAL)" to instead
> > >    "if (IS_ENABLED(CONFIG_SLUB_CPU_PARTIAL))" to fix build error.
> > >    (Thanks Chengming Zhou)
> > >  - Add __maybe_unused for partial_slabs to prevent compiler warning.
> > >
> > > v1:
> > >  https://lore.kernel.org/linux-kernel/20240311132720.37741-1-sxwjean@xxxxxx/T/
> > > ---
> > >  mm/slub.c | 24 +++++++++++++-----------
> > >  1 file changed, 13 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/mm/slub.c b/mm/slub.c
> > > index a3ab096c38c0..ab526960ee5b 100644
> > > --- a/mm/slub.c
> > > +++ b/mm/slub.c
> > > @@ -2588,7 +2588,7 @@ static struct slab *get_partial_node(struct kmem_cache *s,
> > >  {
> > >       struct slab *slab, *slab2, *partial = NULL;
> > >       unsigned long flags;
> > > -     unsigned int partial_slabs = 0;
> > > +     unsigned int __maybe_unused partial_slabs = 0;
> > >
> > >       /*
> > >        * Racy check. If we mistakenly see no partial slabs then we
> > > @@ -2620,19 +2620,21 @@ static struct slab *get_partial_node(struct kmem_cache *s,
> > >               if (!partial) {
> > >                       partial = slab;
> > >                       stat(s, ALLOC_FROM_PARTIAL);
> > > -             } else {
> > > -                     put_cpu_partial(s, slab, 0);
> > > -                     stat(s, CPU_PARTIAL_NODE);
> > > -                     partial_slabs++;
> > > +
> > > +                     /* Fill cpu partial if needed from next iteration, or break */
> > > +                     if (IS_ENABLED(CONFIG_SLUB_CPU_PARTIAL))
> > > +                             continue;
> > > +                     else
> > > +                             break;
> > >               }
> > > -#ifdef CONFIG_SLUB_CPU_PARTIAL
> > > -             if (!kmem_cache_has_cpu_partial(s)
> > > -                     || partial_slabs > s->cpu_partial_slabs / 2)
> > > +
> > > +#if IS_ENABLED(CONFIG_SLUB_CPU_PARTIAL)
> > 
> > Hmm, these two CONFIG_SLUB_CPU_PARTIAL look verbose to me :(
> > 
> > How about using just one, maybe like this?
> > 
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 2ef88bbf56a3..a018c715b648 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -2620,19 +2620,16 @@ static struct slab *get_partial_node(struct kmem_cache *s,
> >                 if (!partial) {
> >                         partial = slab;
> >                         stat(s, ALLOC_FROM_PARTIAL);
> > +#ifdef CONFIG_SLUB_CPU_PARTIAL
> >                 } else {
> 
> I don't like splitting  "if...else..." up with preprocessor directives personally. It's messy for me.
> 

I would agree, but the flow here actually seems like it's easier to follow 
if we are checking for CONFIG_SLUB_CPU_PARTIAL in multiple places :/