Re: [RFC PATCH] mm/slab: save memory by allocating slabobj_ext array from leftover

[Harry Yoo <harry.yoo@oracle.com>]

On Fri, Jun 13, 2025 at 12:42:22PM +0100, Yeoreum Yun wrote:
> Hi Harry,
> 
> [...]

Hi Yeoreum,

> > Allocate slabobj_exts array from this unused space instead of using
> > kcalloc(), when it is large enough.
> >
> > Enjoy the memory savings!
> >
> > [ MEMCG=y, MEM_ALLOC_PROFILING=y ]
> >
> > Before patch (run updatedb):
> >   Slab:            5815196 kB
> >   SReclaimable:    5042824 kB
> >   SUnreclaim:       772372 kB
> >
> > After patch (run updatedb):
> >   Slab:            5748664 kB
> >   SReclaimable:    5041608 kB
> >   SUnreclaim:       707084 kB (-63.75 MiB)
> >
> > [ MEMCG=y, MEM_ALLOC_PROFILING=n ]
> >
> > Before patch (run updatedb):
> >   Slab:            5637764 kB
> >   SReclaimable:    5042428 kB
> >   SUnreclaim:       595284 kB
> >
> > After patch (run updatedb):
> >   Slab:            5598992 kB
> >   SReclaimable:    5042248 kB
> >   SUnreclaim:       560396 kB (-34.07 MiB)
> >
> > This saves from hundreds of KiBs up to several tens of MiBs of memory
> > on my machine, depending on the config and slab memory usage.
> >
> > Enjoy the memory savings!
> 
> Awesome :)

Thanks :)

> [...]
> >  #ifdef CONFIG_SLUB_DEBUG
> >  static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
> >  static DEFINE_SPINLOCK(object_map_lock);
> > @@ -1307,7 +1350,15 @@ slab_pad_check(struct kmem_cache *s, struct slab *slab)
> >       start = slab_address(slab);
> >       length = slab_size(slab);
> >       end = start + length;
> > -     remainder = length % s->size;
> > +
> > +     if (can_alloc_obj_exts_from_leftover(s, slab)) {
> > +             remainder = length;
> > +             remainder -= obj_exts_offset(s, slab);
> > +             remainder -= obj_exts_size(slab);
> > +     } else {
> > +             remainder = length % s->size;
> > +     }
> > +
> >       if (!remainder)
> >               return;
> >
> > @@ -2049,6 +2100,21 @@ static noinline void free_slab_obj_exts(struct slab *slab)
> >       slab->obj_exts = 0;
> >  }
> 
> What concerns me about this patch is the case where !memcg_kmem_online() and
> MEM_ALLOC_PROFILING is not used.
> With this patch, obj_ext can still be created even in that situation,
> and as a result, if data is overwritten in the region previously padded with
> POISON_INUSE (before the patch), slab_pad_check() may no longer catch it

That's a valid point.

I think allocating the array from the leftover space can be deferred
until either MEMCG or MEM_ALLOC_PROFILING actually requests it.

> If this's ignorable, feel free toadd :
> 
> Reviewed-by: Yeoreum Yun <yeoreum.yun@arm.com>

That means the implementation will change a bit, so it's better to drop
the R-b tag as the new change may invalidate "Looks good to me" state.

I'll Cc you in the next version—please take a look and review the
updated version.

Thanks for reviewing!

-- 
Cheers,
Harry / Hyeonggon