Re: [PATCH] slab: remove the SLUB_DEBUG functionality and config option

[David Hildenbrand (Arm)]

On 4/1/26 12:59, Vlastimil Babka (SUSE) wrote:
> The boot-time enabled per-cache debugging has served us well in the
> past, but it's time for it to go, for the following reasons.
> 
> - It's a debugging feature. However, thanks to the advent of LLM-based
>   reviews, we are not adding bugs to the kernel anymore, so it's
>   unnecessary now.

Right, and probably LLM review would find many of bugs in the existing
SLUB_DEBUG code.

> 
> - KASAN is more powerful anyway for the classes of bugs that SLUB_DEBUG
>   can catch. But I suspect KASAN is likely to be removed soon too, see
>   above.

Of course.

> 
> - SLAB never had no such dynamic debugging functionality. With the
>   introduction of sheaves percpu caching, we have turned SLUB back into
>   SLAB partially, so this just follows that direction.
> 
> - It's removing ~2500 lines of code and I want to keep my overal
>   diffstats negative. This adds a nice margin.

Personal stats clearly matter more than anything else.

> 
> Since the slub kunit test depends on SLUB_DEBUG, remove it too. It's not
> a big loss, see the first point above.
> 
> Singed-off-by: Vlastimil Babka (SUSE) <vbabka@xxxxxxxxxx>
> ---
> Signed-off-by: Vlastimil Babka (SUSE) <vbabka@xxxxxxxxxx>
> ---
>  lib/Kconfig.debug      |   12 -
>  lib/Kconfig.kasan      |    2 -
>  lib/tests/Makefile     |    1 -
>  lib/tests/slub_kunit.c |  329 --------
>  mm/Kconfig.debug       |   60 --
>  mm/dmapool.c           |    4 -
>  mm/memcontrol-v1.c     |   17 -
>  mm/mempool.c           |  105 ---
>  mm/slab.h              |   34 -
>  mm/slub.c              | 2074 +-----------------------------------------------

slab.h vs. slub.c is annoying.

Can you just rename that to sheaf.c / sheaf.h now while at it?

-- 
Cheers,

David