Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <f29be74792c7711e0a157a6a024d3998d30be4dd.camel@redhat.com>
Subject: Re: [PATCH] net/core: use xx_zalloc instead xx_alloc and memset
From:   Paolo Abeni <pabeni@redhat.com>
To:     Tian Tao <tiantao6@hisilicon.com>, davem@davemloft.net,
        kuba@kernel.org, linmiaohe@huawei.com, martin.varghese@nokia.com,
        pshelar@ovn.org, fw@strlen.de, viro@zeniv.linux.org.uk,
        gnault@redhat.com, steffen.klassert@secunet.com,
        kyk.segfault@gmail.com, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Wed, 18 Nov 2020 11:47:00 +0100
In-Reply-To: <1605687308-57318-1-git-send-email-tiantao6@hisilicon.com>
References: <1605687308-57318-1-git-send-email-tiantao6@hisilicon.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.36.5 (3.36.5-1.fc32) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Wed, 2020-11-18 at 16:15 +0800, Tian Tao wrote:
> use kmem_cache_zalloc instead kmem_cache_alloc and memset.
> 
> Signed-off-by: Tian Tao <tiantao6@hisilicon.com>
> ---
>  net/core/skbuff.c | 10 +++-------
>  1 file changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index c9a5a3c..3449c1c 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -313,12 +313,10 @@ struct sk_buff *__build_skb(void *data, unsigned int frag_size)
>  {
>  	struct sk_buff *skb;
>  
> -	skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
> +	skb = kmem_cache_zalloc(skbuff_head_cache, GFP_ATOMIC);

This will zeroed a slighly larger amount of data compared to the
existing code: offsetof(struct sk_buff, tail) == 182, sizeof(struct
sk_buff) == 224.

>  	if (unlikely(!skb))
>  		return NULL;
>  
> -	memset(skb, 0, offsetof(struct sk_buff, tail));

Additionally this leverages constant argument optimizations.

Possibly overall not noticeable, but this code path is quite critical
performance wise.

I would avoid the above.
> -
>  	return __build_skb_around(skb, data, frag_size);
>  }
>  
> @@ -6170,12 +6168,10 @@ static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
>   */
>  struct skb_ext *__skb_ext_alloc(gfp_t flags)
>  {
> -	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
> +	struct skb_ext *new = kmem_cache_zalloc(skbuff_ext_cache, flags);
>  
> -	if (new) {
> -		memset(new->offset, 0, sizeof(new->offset));

Similar to the above, but additionally here the number of zeroed bytes
changes a lot and a few additional cachelines will be touched. The
performance impact is likely relevant.

Overall I think we should not do this.

Thanks,

Paolo