Re: [PATCH v5 02/16] mm: introduce kmem_cache_alloc_lru

[Roman Gushchin <guro@fb.com>]

On Tue, Jan 11, 2022 at 04:41:29PM +0100, Vlastimil Babka wrote:
> On 1/10/22 19:47, Roman Gushchin wrote:
> > On Sun, Jan 09, 2022 at 02:21:22PM +0800, Muchun Song wrote:
> >> On Fri, Jan 7, 2022 at 11:05 AM Roman Gushchin <guro@fb.com> wrote:
> >> >
> >> [...]
> >> > >  /*
> >> > >   * struct kmem_cache related prototypes
> >> > > @@ -425,6 +426,8 @@ static __always_inline unsigned int __kmalloc_index(size_t size,
> >> > >
> >> > >  void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __alloc_size(1);
> >> > >  void *kmem_cache_alloc(struct kmem_cache *s, gfp_t flags) __assume_slab_alignment __malloc;
> >> > > +void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
> >> > > +                        gfp_t gfpflags) __assume_slab_alignment __malloc;
> >> >
> >> > I'm not a big fan of this patch: I don't see why preparing the lru
> >> > infrastructure has to be integrated that deep into the slab code.
> >> >
> >> > Why can't kmem_cache_alloc_lru() be a simple wrapper like (pseudo-code):
> >> >   void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
> >> >                            gfp_t gfpflags) {
> >> >         if (necessarily)
> >> >            prepare_lru_infra();
> >> >         return kmem_cache_alloc();
> >> >   }
> >> 
> >> Hi Roman,
> >> 
> >> Actually, it can. But there is going to be some redundant code similar
> >> like memcg_slab_pre_alloc_hook() does to detect the necessity of
> >> prepare_lru_infra() in the new scheme of kmem_cache_alloc_lru().
> >> I just want to reduce the redundant overhead.
> > 
> > Is this about getting a memcg pointer?
> > I doubt it's a good reason to make changes all over the slab code.
> > Another option to consider adding a new gfp flag.
> 
> I'm not sure how a flag would help as it seems we really need to pass a
> specific list_lru pointer and work with that. I was thinking if there was
> only one list_lru per class of object it could be part of struct kmem_cache,
> but investigating kmem_cache_alloc_lru() callers I see lru parameters:
> 
> - &nfs4_xattr_cache_lru - this is fixed
> - xas->xa_lru potentially not fixed, although the only caller of
> xas_set_lru() passes &shadow_nodes so effectively fixed
> - &sb->s_dentry_lru - dynamic, boo

Indeed.

> 
> > Vlastimil, what do you think?
> 
> Memcg code is already quite intertwined with slab code, for better or worse,
> so I guess the extra lru parameter in a bunch of inline functions won't
> change much. I don't immediately see a better solution.

Ok then. Thanks for taking a look!