Re: [PATCH] docs: ABI: sysfs-kernel-slab: mark cpu_partial attributes deprecated

["Vlastimil Babka (SUSE)" <vbabka@kernel.org>]

On 7/2/26 07:18, Harry Yoo wrote:
> 
> 
> On 7/1/26 11:17 PM, Seongjun Hong wrote:
>> The per-cpu slab and per-cpu partial slab mechanisms were removed when
>> SLUB was converted to per-cpu sheaves in Linux 7.0. The cpu_slabs,
> 
> small nit: sheaves were introduced in v6.18,
> but the cpu (partial) slabs layer were removed later in v7.0.

Fixed up when applying by saying "fully converted" :)

> 
>> slabs_cpu_partial and cpu_partial sysfs attributes were kept as stubs
>> that always return 0 for backwards compatibility, but their
>> documentation still described them as if they were functional.
> 
> Right, for backward compatibility, because, unlike files that
> might not exist w/o SLUB_STATS, userspace tools might assume that
> they always exist.
> 
>> Update the three descriptions to state that the attributes are
>> deprecated and always read 0, and note that they are retained only for
>> compatibility. While here, fix a "partialli" typo in the
>> slabs_cpu_partial description.
>> 
>> Signed-off-by: Seongjun Hong <hsj0512@snu.ac.kr>
>> ---
> 
> Overall looks good to me, so:
> Acked-by: Harry Yoo (Oracle) <harry@kernel.org>

Added to slab/for-next, thanks!

> By the way.
> 
> There's still some gap between the documentation
> and the implementation :)
> 
> We've dropped a bunch of files that exist only on
> SLUB_STATS and new files introduced with by
> sheaves were not documented.

While we are touching these files, the question is what contact to use
there. Some have Christoph, some Pekka, some both.
Seems inflexible to me and should ideally point to the slab MAINTAINERS
entry. But not sure if it's possible. Documentation/ABI/README says:

Contact:        Primary contact for this interface (may be a mailing list)

Maybe we can propose that it can also say

See the SLAB ALLOCATOR subsystem in MAINTAINERS.

> Also, tools/mm/slabinfo.c needs some attention as
> it hasn't been updated for a while.

There were some fixes in 7.2, but perhaps not enough.

> Thanks.
>