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;
Date:   Tue, 16 Feb 2021 11:39:21 +0000
To:     Magnus Karlsson <magnus.karlsson@intel.com>,
        =?utf-8?Q?Bj=C3=B6rn_T=C3=B6pel?= <bjorn@kernel.org>
From:   Alexander Lobakin <alobakin@pm.me>
Cc:     "Michael S. Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>,
        "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Jonathan Lemon <jonathan.lemon@gmail.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Jesper Dangaard Brouer <hawk@kernel.org>,
        John Fastabend <john.fastabend@gmail.com>,
        Andrii Nakryiko <andrii@kernel.org>,
        Martin KaFai Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
        KP Singh <kpsingh@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
        Eric Dumazet <eric.dumazet@gmail.com>,
        Xuan Zhuo <xuanzhuo@linux.alibaba.com>,
        Dust Li <dust.li@linux.alibaba.com>,
        Alexander Lobakin <alobakin@pm.me>,
        virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, bpf@vger.kernel.org
Reply-To: Alexander Lobakin <alobakin@pm.me>
Subject: [PATCH v4 bpf-next 6/6] xsk: build skb by page (aka generic zerocopy xmit)
Message-ID: <20210216113740.62041-7-alobakin@pm.me>
In-Reply-To: <20210216113740.62041-1-alobakin@pm.me>
References: <20210216113740.62041-1-alobakin@pm.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

From: Xuan Zhuo <xuanzhuo@linux.alibaba.com>

This patch is used to construct skb based on page to save memory copy
overhead.

This function is implemented based on IFF_TX_SKB_NO_LINEAR. Only the
network card priv_flags supports IFF_TX_SKB_NO_LINEAR will use page to
directly construct skb. If this feature is not supported, it is still
necessary to copy data to construct skb.

---------------- Performance Testing ------------

The test environment is Aliyun ECS server.
Test cmd:
```
xdpsock -i eth0 -t  -S -s <msg size>
```

Test result data:

size    64      512     1024    1500
copy    1916747 1775988 1600203 1440054
page    1974058 1953655 1945463 1904478
percent 3.0%    10.0%   21.58%  32.3%

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Reviewed-by: Dust Li <dust.li@linux.alibaba.com>
[ alobakin:
 - expand subject to make it clearer;
 - improve skb->truesize calculation;
 - reserve some headroom in skb for drivers;
 - tailroom is not needed as skb is non-linear ]
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
---
 net/xdp/xsk.c | 119 ++++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 95 insertions(+), 24 deletions(-)

diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 143979ea4165..ff7bd06e1241 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -445,6 +445,96 @@ static void xsk_destruct_skb(struct sk_buff *skb)
 =09sock_wfree(skb);
 }
=20
+static struct sk_buff *xsk_build_skb_zerocopy(struct xdp_sock *xs,
+=09=09=09=09=09      struct xdp_desc *desc)
+{
+=09struct xsk_buff_pool *pool =3D xs->pool;
+=09u32 hr, len, offset, copy, copied;
+=09struct sk_buff *skb;
+=09struct page *page;
+=09void *buffer;
+=09int err, i;
+=09u64 addr;
+
+=09hr =3D max(NET_SKB_PAD, L1_CACHE_ALIGN(xs->dev->needed_headroom));
+
+=09skb =3D sock_alloc_send_skb(&xs->sk, hr, 1, &err);
+=09if (unlikely(!skb))
+=09=09return ERR_PTR(err);
+
+=09skb_reserve(skb, hr);
+
+=09addr =3D desc->addr;
+=09len =3D desc->len;
+
+=09buffer =3D xsk_buff_raw_get_data(pool, addr);
+=09offset =3D offset_in_page(buffer);
+=09addr =3D buffer - pool->addrs;
+
+=09for (copied =3D 0, i =3D 0; copied < len; i++) {
+=09=09page =3D pool->umem->pgs[addr >> PAGE_SHIFT];
+=09=09get_page(page);
+
+=09=09copy =3D min_t(u32, PAGE_SIZE - offset, len - copied);
+=09=09skb_fill_page_desc(skb, i, page, offset, copy);
+
+=09=09copied +=3D copy;
+=09=09addr +=3D copy;
+=09=09offset =3D 0;
+=09}
+
+=09skb->len +=3D len;
+=09skb->data_len +=3D len;
+=09skb->truesize +=3D pool->unaligned ? len : pool->chunk_size;
+
+=09refcount_add(skb->truesize, &xs->sk.sk_wmem_alloc);
+
+=09return skb;
+}
+
+static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
+=09=09=09=09     struct xdp_desc *desc)
+{
+=09struct net_device *dev =3D xs->dev;
+=09struct sk_buff *skb;
+
+=09if (dev->priv_flags & IFF_TX_SKB_NO_LINEAR) {
+=09=09skb =3D xsk_build_skb_zerocopy(xs, desc);
+=09=09if (IS_ERR(skb))
+=09=09=09return skb;
+=09} else {
+=09=09u32 hr, tr, len;
+=09=09void *buffer;
+=09=09int err;
+
+=09=09hr =3D max(NET_SKB_PAD, L1_CACHE_ALIGN(dev->needed_headroom));
+=09=09tr =3D dev->needed_tailroom;
+=09=09len =3D desc->len;
+
+=09=09skb =3D sock_alloc_send_skb(&xs->sk, hr + len + tr, 1, &err);
+=09=09if (unlikely(!skb))
+=09=09=09return ERR_PTR(err);
+
+=09=09skb_reserve(skb, hr);
+=09=09skb_put(skb, len);
+
+=09=09buffer =3D xsk_buff_raw_get_data(xs->pool, desc->addr);
+=09=09err =3D skb_store_bits(skb, 0, buffer, len);
+=09=09if (unlikely(err)) {
+=09=09=09kfree_skb(skb);
+=09=09=09return ERR_PTR(err);
+=09=09}
+=09}
+
+=09skb->dev =3D dev;
+=09skb->priority =3D xs->sk.sk_priority;
+=09skb->mark =3D xs->sk.sk_mark;
+=09skb_shinfo(skb)->destructor_arg =3D (void *)(long)desc->addr;
+=09skb->destructor =3D xsk_destruct_skb;
+
+=09return skb;
+}
+
 static int xsk_generic_xmit(struct sock *sk)
 {
 =09struct xdp_sock *xs =3D xdp_sk(sk);
@@ -454,56 +544,37 @@ static int xsk_generic_xmit(struct sock *sk)
 =09struct sk_buff *skb;
 =09unsigned long flags;
 =09int err =3D 0;
-=09u32 hr, tr;
=20
 =09mutex_lock(&xs->mutex);
=20
 =09if (xs->queue_id >=3D xs->dev->real_num_tx_queues)
 =09=09goto out;
=20
-=09hr =3D max(NET_SKB_PAD, L1_CACHE_ALIGN(xs->dev->needed_headroom));
-=09tr =3D xs->dev->needed_tailroom;
-
 =09while (xskq_cons_peek_desc(xs->tx, &desc, xs->pool)) {
-=09=09char *buffer;
-=09=09u64 addr;
-=09=09u32 len;
-
 =09=09if (max_batch-- =3D=3D 0) {
 =09=09=09err =3D -EAGAIN;
 =09=09=09goto out;
 =09=09}
=20
-=09=09len =3D desc.len;
-=09=09skb =3D sock_alloc_send_skb(sk, hr + len + tr, 1, &err);
-=09=09if (unlikely(!skb))
+=09=09skb =3D xsk_build_skb(xs, &desc);
+=09=09if (IS_ERR(skb)) {
+=09=09=09err =3D PTR_ERR(skb);
 =09=09=09goto out;
+=09=09}
=20
-=09=09skb_reserve(skb, hr);
-=09=09skb_put(skb, len);
-
-=09=09addr =3D desc.addr;
-=09=09buffer =3D xsk_buff_raw_get_data(xs->pool, addr);
-=09=09err =3D skb_store_bits(skb, 0, buffer, len);
 =09=09/* This is the backpressure mechanism for the Tx path.
 =09=09 * Reserve space in the completion queue and only proceed
 =09=09 * if there is space in it. This avoids having to implement
 =09=09 * any buffering in the Tx path.
 =09=09 */
 =09=09spin_lock_irqsave(&xs->pool->cq_lock, flags);
-=09=09if (unlikely(err) || xskq_prod_reserve(xs->pool->cq)) {
+=09=09if (xskq_prod_reserve(xs->pool->cq)) {
 =09=09=09spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
 =09=09=09kfree_skb(skb);
 =09=09=09goto out;
 =09=09}
 =09=09spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
=20
-=09=09skb->dev =3D xs->dev;
-=09=09skb->priority =3D sk->sk_priority;
-=09=09skb->mark =3D sk->sk_mark;
-=09=09skb_shinfo(skb)->destructor_arg =3D (void *)(long)desc.addr;
-=09=09skb->destructor =3D xsk_destruct_skb;
-
 =09=09err =3D __dev_direct_xmit(skb, xs->queue_id);
 =09=09if  (err =3D=3D NETDEV_TX_BUSY) {
 =09=09=09/* Tell user-space to retry the send */
--=20
2.30.1