From: Jesper Dangaard Brouer <hawk@xxxxxxxxxx>
Date: Tue, 15 Aug 2023 17:17:36 +0200
> Since v6.5-rc1 MM-tree is merged and contains a new flag SLAB_NO_MERGE
> in commit d0bf7d5759c1 ("mm/slab: introduce kmem_cache flag SLAB_NO_MERGE")
> now is the time to use this flag for networking as proposed
> earlier see link.
>
> The SKB (sk_buff) kmem_cache slab is critical for network performance.
> Network stack uses kmem_cache_{alloc,free}_bulk APIs to gain
> performance by amortising the alloc/free cost.
>
> For the bulk API to perform efficiently the slub fragmentation need to
> be low. Especially for the SLUB allocator, the efficiency of bulk free
> API depend on objects belonging to the same slab (page).
>
> When running different network performance microbenchmarks, I started
> to notice that performance was reduced (slightly) when machines had
> longer uptimes. I believe the cause was 'skbuff_head_cache' got
> aliased/merged into the general slub for 256 bytes sized objects (with
> my kernel config, without CONFIG_HARDENED_USERCOPY).
>
> For SKB kmem_cache network stack have other various reasons for
> not merging, but it varies depending on kernel config (e.g.
> CONFIG_HARDENED_USERCOPY). We want to explicitly set SLAB_NO_MERGE
> for this kmem_cache to get most out of kmem_cache_{alloc,free}_bulk APIs.
>
> When CONFIG_SLUB_TINY is configured the bulk APIs are essentially
> disabled. Thus, for this case drop the SLAB_NO_MERGE flag.
>
> Link: https://lore.kernel.org/all/167396280045.539803.7540459812377220500.stgit@firesoul/
> Signed-off-by: Jesper Dangaard Brouer <hawk@xxxxxxxxxx>
> ---
> net/core/skbuff.c | 13 ++++++++++++-
> 1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index a298992060e6..92aee3e0376a 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -4750,12 +4750,23 @@ static void skb_extensions_init(void)
> static void skb_extensions_init(void) {}
> #endif
>
> +/* The SKB kmem_cache slab is critical for network performance. Never
> + * merge/alias the slab with similar sized objects. This avoids fragmentation
> + * that hurts performance of kmem_cache_{alloc,free}_bulk APIs.
> + */
> +#ifndef CONFIG_SLUB_TINY
> +#define FLAG_SKB_NO_MERGE SLAB_NO_MERGE
> +#else /* CONFIG_SLUB_TINY - simple loop in kmem_cache_alloc_bulk */
> +#define FLAG_SKB_NO_MERGE 0
> +#endif
> +
> void __init skb_init(void)
> {
> skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
> sizeof(struct sk_buff),
> 0,
> - SLAB_HWCACHE_ALIGN|SLAB_PANIC,
> + SLAB_HWCACHE_ALIGN|SLAB_PANIC|
> + FLAG_SKB_NO_MERGE,
That alignment tho xD
> offsetof(struct sk_buff, cb),
> sizeof_field(struct sk_buff, cb),
> NULL);
Thanks,
Olek
