Re: [PATCH v2 12/13] mm/slab: add cache_ and slab_needs_objcg() helpers

On Fri, Jul 24, 2026 at 06:31:51PM +0200, Vlastimil Babka (SUSE) wrote:
> On 7/24/26 11:41, Hao Li wrote:
> > On Mon, Jul 20, 2026 at 04:16:26PM +0200, Vlastimil Babka (SUSE) wrote:
> >> Slabs of some caches never need the objcg part of struct slabobj_ext.
> >> Introduce helpers to query this for a cache or a slab.
> >> 
> >> Introduce SLAB_MAY_ACCOUNT flag that is currently only internal and only
> >> is_kmalloc_normal() caches don't have the flag.
> > 
> > it seems this could be "only type == KMALLOC_NORMAL caches don't have this
> > flag".
> 
> I think not, because with SLUB_TINY,

Ah, right, I missed the SLUB_TINY case.

> a KMALLOC_NORMAL cache can have
> SLAB_RECLAIM_ACCOUNT flag and thus is not is_kmalloc_normal()

Just a minor detail though: under SLUB_TINY, SLAB_RECLAIM_ACCOUNT isn't used,
and new_kmalloc_cache() checks KMALLOC_RECLAIM != KMALLOC_NORMAL before setting
it. So is_kmalloc_normal() should still evaluate to true.

> and may need objcgs?

Agree. Under SLUB_TINY, we indeed need to set SLAB_MAY_ACCOUNT.

> 
> Yep it's subtle :/
> 

Yeah, it's pretty subtle. Since the full case matrix is tricky to summarize
accurately in one sentence, I think it might be cleaner to simply drop "only
is_kmalloc_normal() caches don't have the flag".

-- 
Thanks,
Hao