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;
MIME-Version: 1.0
References: <20210312162127.239795-1-alobakin@pm.me> <20210312162127.239795-3-alobakin@pm.me>
In-Reply-To: <20210312162127.239795-3-alobakin@pm.me>
From:   Eric Dumazet <edumazet@google.com>
Date:   Fri, 12 Mar 2021 17:47:04 +0100
Message-ID: <CANn89iJvmakwfLFb6QS1ettiJM-D3cJ89bFPvZ=Gk2YyGpxQuw@mail.gmail.com>
Subject: Re: [PATCH net-next 2/4] gro: don't dereference napi->gro_hash[x]
 multiple times in dev_gro_receive()
To:     Alexander Lobakin <alobakin@pm.me>
Cc:     "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andriin@fb.com>, Wei Wang <weiwan@google.com>,
        Cong Wang <cong.wang@bytedance.com>,
        Taehee Yoo <ap420073@gmail.com>,
        netdev <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Mar 12, 2021 at 5:22 PM Alexander Lobakin <alobakin@pm.me> wrote:
>
> GRO bucket index doesn't change through the entire function.
> Store a pointer to the corresponding bucket on stack once and use
> it later instead of dereferencing again and again.
>
> Signed-off-by: Alexander Lobakin <alobakin@pm.me>
> ---
>  net/core/dev.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index adc42ba7ffd8..ee124aecb8a2 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -5957,6 +5957,7 @@ static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
>  static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
>  {
>         u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
> +       struct gro_list *gro_list = &napi->gro_hash[bucket];
>         struct list_head *head = &offload_base;
>         struct packet_offload *ptype;
>         __be16 type = skb->protocol;
> @@ -6024,7 +6025,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
>         if (pp) {
>                 skb_list_del_init(pp);
>                 napi_gro_complete(napi, pp);
> -               napi->gro_hash[bucket].count--;
> +               gro_list->count--;
>         }
>
>         if (same_flow)
> @@ -6033,10 +6034,10 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
>         if (NAPI_GRO_CB(skb)->flush)
>                 goto normal;
>
> -       if (unlikely(napi->gro_hash[bucket].count >= MAX_GRO_SKBS)) {
> +       if (unlikely(gro_list->count >= MAX_GRO_SKBS)) {
>                 gro_flush_oldest(napi, gro_head);
>         } else {
> -               napi->gro_hash[bucket].count++;
> +               gro_list->count++;
>         }
>         NAPI_GRO_CB(skb)->count = 1;
>         NAPI_GRO_CB(skb)->age = jiffies;
> @@ -6050,7 +6051,7 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
>         if (grow > 0)
>                 gro_pull_from_frag0(skb, grow);
>  ok:
> -       if (napi->gro_hash[bucket].count) {
> +       if (gro_list->count) {
>                 if (!test_bit(bucket, &napi->gro_bitmask))
>                         __set_bit(bucket, &napi->gro_bitmask);
>         } else if (test_bit(bucket, &napi->gro_bitmask)) {
> --
> 2.30.2
>
>

This adds more register pressure, do you have precise measures to
confirm this change is a win ?

Presumably the compiler should be able to optimize the code just fine,
it can see @bucket does not change.