Re: [PATCH v2] mm/slub: free prefilled empty sheaves consistently

[hu.shengming]

Hao wrote:
> On Fri, May 22, 2026 at 02:59:00PM +0800, hu.shengming@xxxxxxxxxx wrote:
> > From: Shengming Hu <hu.shengming@xxxxxxxxxx>
> > 
> > Empty sheaves allocated by alloc_empty_sheaf() should be released with
> > free_empty_sheaf().
> > 
> > Move the common empty-sheaf initialization into __alloc_empty_sheaf(),
> > including the capacity and default pfmemalloc state, and use it for
> > oversized prefilled sheaves as well.
> > 
> > This keeps the prefill paths consistent: after flushing an oversized or
> > pfmemalloc sheaf, release the now-empty sheaf through free_empty_sheaf()
> > instead of kfree().
> > 
> > Signed-off-by: Shengming Hu <hu.shengming@xxxxxxxxxx>
> > ---
> > Changes in v2:
> > - Rework the change as suggested by Harry.
> > - Teach __alloc_empty_sheaf() to initialize capacity and pfmemalloc.
> > - Allocate oversized prefilled sheaves through __alloc_empty_sheaf().
> > - Free flushed oversized and pfmemalloc sheaves through free_empty_sheaf().
> > - Link to v1: https://lore.kernel.org/all/20260521195015105Y4zvKHj0TfPZEujixy9Vo@xxxxxxxxxx/
> > 
> > ---
> >  mm/slub.c | 17 +++++------------
> >  1 file changed, 5 insertions(+), 12 deletions(-)
> > 
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 04692a6f9128..bbe8351f1bfa 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -2782,6 +2782,8 @@ static struct slab_sheaf *__alloc_empty_sheaf(struct kmem_cache *s, gfp_t gfp,
> >          return NULL;
> > 
> >      sheaf->cache = s;
> > +    sheaf->capacity = capacity;
> > +    sheaf->pfmemalloc = false;
> 
> we could eliminate `sheaf->pfmemalloc = false;` because the allocation use
> kzalloc.
> 

Hi Hao,

Thanks for the review.

Yes, I agree that `sheaf->pfmemalloc = false` in
__alloc_empty_sheaf() is redundant, as the sheaf is allocated through
kzalloc().

> > 
> >      stat(s, SHEAF_ALLOC);
> > 
> > @@ -5015,21 +5017,15 @@ kmem_cache_prefill_sheaf(struct kmem_cache *s, gfp_t gfp, unsigned int size)
> > 
> >      if (unlikely(size > s->sheaf_capacity)) {
> > 
> > -        sheaf = kzalloc_flex(*sheaf, objects, size, gfp);
> > +        sheaf = __alloc_empty_sheaf(s, gfp, size);
> >          if (!sheaf)
> >              return NULL;
> > 
> >          stat(s, SHEAF_PREFILL_OVERSIZE);
> > -        sheaf->cache = s;
> > -        sheaf->capacity = size;
> > 
> > -        /*
> > -         * we do not need to care about pfmemalloc here because oversize
> > -         * sheaves area always flushed and freed when returned
> > -         */
> >          if (!__kmem_cache_alloc_bulk(s, gfp, size,
> >                           &sheaf->objects[0])) {
> > -            kfree(sheaf);
> > +            free_empty_sheaf(s, sheaf);
> >              return NULL;
> >          }
> > 
> > @@ -5064,9 +5060,6 @@ kmem_cache_prefill_sheaf(struct kmem_cache *s, gfp_t gfp, unsigned int size)
> >          sheaf = alloc_empty_sheaf(s, gfp);
> > 
> >      if (sheaf) {
> > -        sheaf->capacity = s->sheaf_capacity;
> > -        sheaf->pfmemalloc = false;
> 
> I think we couldn't delete 'sheaf->pfmemalloc = false;', because the 'sheaf'
> pointer here could points to a normal size sheaf via the `if (pcs->spare) {}
> else {}` statements above.
> 

You are also right that sheaf in the normal-sized prefill path may come
from pcs->spare or the barn rather than from a fresh
__alloc_empty_sheaf() allocation. Therefore, the pfmemalloc state
cannot rely on the zeroed allocation there, and I will restore the
explicit initialization in kmem_cache_prefill_sheaf().

Also, after looking at this again, I found that capacity needs the same
treatment.`capacity` and `pfmemalloc` share storage with `rcu_head`/`barn_list`,
so the prefill-specific fields need to be reinitialized whenever a normal-sized
sheaf is reused as a prefilled sheaf.

I will restore both assignments in v3.

> > -
> >          if (sheaf->size < size &&
> >              __prefill_sheaf_pfmemalloc(s, sheaf, gfp)) {
> >              sheaf_flush_unused(s, sheaf);
> > @@ -5097,7 +5090,7 @@ void kmem_cache_return_sheaf(struct kmem_cache *s, gfp_t gfp,
> >      if (unlikely((sheaf->capacity != s->sheaf_capacity)
> >               || sheaf->pfmemalloc)) {
> >          sheaf_flush_unused(s, sheaf);
> > -        kfree(sheaf);
> > +        free_empty_sheaf(s, sheaf);
> >          return;
> >      }
> > 
> > -- 
> > 2.25.1
> 
> -- 
> Thanks,
> Hao

--
With Best Regards,
Shengming