Re: [PATCH 05/12] mm, slub: lower the default slub_max_order with CONFIG_SLUB_TINY

From: Hyeonggon Yoo <42.hyeyoo@gmail.com>
To: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>,
	David Rientjes <rientjes@google.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	Pekka Enberg <penberg@kernel.org>,
	Roman Gushchin <roman.gushchin@linux.dev>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Matthew Wilcox <willy@infradead.org>,
	patches@lists.linux.dev, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 05/12] mm, slub: lower the default slub_max_order with CONFIG_SLUB_TINY
Date: Thu, 24 Nov 2022 20:40:55 +0900	[thread overview]
Message-ID: <Y39YR5nn6aUs2KRW@hyeyoo> (raw)
In-Reply-To: <20221121171202.22080-6-vbabka@suse.cz>

On Mon, Nov 21, 2022 at 06:11:55PM +0100, Vlastimil Babka wrote:
> With CONFIG_SLUB_TINY we want to minimize memory overhead. By lowering
> the default slub_max_order we can make slab allocations use smaller
> pages. However depending on object sizes, order-0 might not be the best
> due to increased fragmentation. When testing on a 8MB RAM k210 system by
> Damien Le Moal [1], slub_max_order=1 had the best results, so use that
> as the default for CONFIG_SLUB_TINY.
> 
> [1] https://lore.kernel.org/all/6a1883c4-4c3f-545a-90e8-2cd805bcf4ae@opensource.wdc.com/
> 
> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
> ---
>  mm/slub.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 917b79278bad..bf726dd00f7d 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -3888,7 +3888,8 @@ EXPORT_SYMBOL(kmem_cache_alloc_bulk);
>   * take the list_lock.
>   */
>  static unsigned int slub_min_order;
> -static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
> +static unsigned int slub_max_order =
> +	IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER;
>  static unsigned int slub_min_objects;
>  
>  /*
> -- 
> 2.38.1
>

Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>

-- 
Thanks,
Hyeonggon