Received: by 2002:a05:6a10:8c0a:0:0:0:0 with SMTP id go10csp2468797pxb; Mon, 18 Jan 2021 20:14:09 -0800 (PST) X-Google-Smtp-Source: ABdhPJzm5mYGu2YcfdI6RwXXcC7GKBrH+xjWFUCkoP1MmzitoACd4dmuIP32UC5P4IWlLJ2YI//I X-Received: by 2002:a17:907:2458:: with SMTP id yw24mr1763482ejb.333.1611029649524; Mon, 18 Jan 2021 20:14:09 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1611029649; cv=none; d=google.com; s=arc-20160816; b=i/sn1KxCxu4fzrWHApDcPUZUJrnCVVnsSXcd7WCfpPW8gFe+/GytJWSHM/cAkmda64 1LSaOVKVqrlQ/aebVgFABP83GNMuNEmOSlFr/ZHQ2sfWFaZvuy6/X0hC4SHj02TuT55b vtL/CoZtFkLXa8Vv7hBx/p1VyscjozFnMO+lVjTPA+/M8pXy9IAlDfUZOf32FYBWbv2p +oYDfLwvbPInYZ2/wHfp9/ZLsD++op/VkwH2GuA9YBNJ5LgRYXLXrb5s2ayqeiaNExS7 Dr7PRj2Fy6PNEzi3LZeXgd7uTk9+ebUCFsFBXG7k4NP/ecmgHORAHW25U2SEdgHkY8GI BdPA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=list-id:precedence:content-transfer-encoding:mime-version :references:in-reply-to:message-id:subject:reply-to:cc:from:to :dkim-signature:date; bh=0Xcw0aAhmjtsp5cwUHHVIctjC2VS/St8x2ljm1ElZ08=; b=D16dLsdT+mpl2f4qwiIkxjl4P6hGGqGG3JpVG7/o53pC4p5LwBw1HuQcJ3Y7lTJjeO qpvo560qcCh2NW2Z7SgboOILq0xhBhXtGydz84okKjHlXOI3BFyFKQAy4bbyNErqX87v Zj6b2JjRbmgHlsRxi3o/gFrW14aB9ZX/ASU7NCMTS5ahAHBL5zddELYCDXquUgZUv/Iw 7R41WVFyxNTpT199s9tRY/OOQSttSny9gTif42bZpU9/tdJ2xL92yGlL/cSiX4qisev1 fpS2BF+pOFMezvB43plZsvhXgpVuUsbPRI1OiYByPZ0TVG3ROb8fKK/n4UN+eiXLVo3T nw1g== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@pm.me header.s=protonmail header.b=ZYwW6BVB; spf=pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) smtp.mailfrom=linux-kernel-owner@vger.kernel.org; dmarc=pass (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=pm.me Return-Path: Received: from vger.kernel.org (vger.kernel.org. [23.128.96.18]) by mx.google.com with ESMTP id u22si7709669ejc.544.2021.01.18.20.13.46; Mon, 18 Jan 2021 20:14:09 -0800 (PST) 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; Authentication-Results: mx.google.com; dkim=pass header.i=@pm.me header.s=protonmail header.b=ZYwW6BVB; spf=pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) smtp.mailfrom=linux-kernel-owner@vger.kernel.org; dmarc=pass (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=pm.me Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S2404245AbhARMnU (ORCPT + 99 others); Mon, 18 Jan 2021 07:43:20 -0500 Received: from mail-40133.protonmail.ch ([185.70.40.133]:20400 "EHLO mail-40133.protonmail.ch" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S2404064AbhARMiK (ORCPT ); Mon, 18 Jan 2021 07:38:10 -0500 Date: Mon, 18 Jan 2021 12:37:17 +0000 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=pm.me; s=protonmail; t=1610973443; bh=0Xcw0aAhmjtsp5cwUHHVIctjC2VS/St8x2ljm1ElZ08=; h=Date:To:From:Cc:Reply-To:Subject:In-Reply-To:References:From; b=ZYwW6BVBSXvSTbxxWWc3tWy00U1q4AmGSsBiofpx54va08p1NdoME33vPRXdMM/i/ XmsIQXFwzpj6ppGvemwavlNYcHuNIdYgMbnm8j/5VB+R04m7ZQcX6AGRP/GU8DoYRB ObSNvSPQVe4kAihx7vTnq6mALbfXJg6mzHUssgFygsLRZNN/VnhweHBObjRiVIN7kz 5XebaYgRimuNlnUlJRTuE/I44jRIUcMKfSh6JUEOG2UOJEJ3Y235zDrO4V0rp6xAsh HgHQEya+7jJiio3oROuxcvIbJMYWUvI2UjIMlMaIvwNPiFbmudG5VAmCgOdSK7LZo8 KGSg8ySXY2G2A== To: Xuan Zhuo From: Alexander Lobakin Cc: Alexander Lobakin , "Michael S. Tsirkin" , Jason Wang , "David S. Miller" , Jakub Kicinski , bjorn.topel@intel.com, Magnus Karlsson , Jonathan Lemon , Alexei Starovoitov , Daniel Borkmann , Jesper Dangaard Brouer , John Fastabend , Andrii Nakryiko , Martin KaFai Lau , Song Liu , Yonghong Song , KP Singh , Willem de Bruijn , Steffen Klassert , Miaohe Lin , Mauro Carvalho Chehab , Antoine Tenart , Michal Kubecek , Andrew Lunn , Florian Fainelli , Meir Lichtinger , virtualization@lists.linux-foundation.org, bpf@vger.kernel.org, netdev@vger.kernel.org, linux-kernel@vger.kernel.org Reply-To: Alexander Lobakin Subject: Re: [PATCH bpf-next] xsk: build skb by page Message-ID: <20210118123645.2635-1-alobakin@pm.me> In-Reply-To: <579fa463bba42ac71591540a1811dca41d725350.1610764948.git.xuanzhuo@linux.alibaba.com> References: <579fa463bba42ac71591540a1811dca41d725350.1610764948.git.xuanzhuo@linux.alibaba.com> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-1.2 required=10.0 tests=ALL_TRUSTED,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF shortcircuit=no autolearn=disabled version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on mailout.protonmail.ch Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org From: Xuan Zhuo Date: Sat, 16 Jan 2021 10:44:53 +0800 > This patch is used to construct skb based on page to save memory copy > overhead. >=20 > This has one problem: >=20 > We construct the skb by fill the data page as a frag into the skb. In > this way, the linear space is empty, and the header information is also > in the frag, not in the linear space, which is not allowed for some > network cards. For example, Mellanox Technologies MT27710 Family > [ConnectX-4 Lx] will get the following error message: >=20 > mlx5_core 0000:3b:00.1 eth1: Error cqe on cqn 0x817, ci 0x8, qn 0x1db= b, opcode 0xd, syndrome 0x1, vendor syndrome 0x68 > 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > 00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > 00000030: 00 00 00 00 60 10 68 01 0a 00 1d bb 00 0f 9f d2 > WQE DUMP: WQ size 1024 WQ cur size 0, WQE index 0xf, len: 64 > 00000000: 00 00 0f 0a 00 1d bb 03 00 00 00 08 00 00 00 00 > 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > 00000020: 00 00 00 2b 00 08 00 00 00 00 00 05 9e e3 08 00 > 00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 > mlx5_core 0000:3b:00.1 eth1: ERR CQE on SQ: 0x1dbb >=20 > I also tried to use build_skb to construct skb, but because of the > existence of skb_shinfo, it must be behind the linear space, so this > method is not working. We can't put skb_shinfo on desc->addr, it will be > exposed to users, this is not safe. >=20 > Finally, I added a feature NETIF_F_SKB_NO_LINEAR to identify whether the > network card supports the header information of the packet in the frag > and not in the linear space. >=20 > ---------------- Performance Testing ------------ >=20 > The test environment is Aliyun ECS server. > Test cmd: > ``` > xdpsock -i eth0 -t -S -s > ``` >=20 > Test result data: >=20 > size 64 512 1024 1500 > copy 1916747 1775988 1600203 1440054 > page 1974058 1953655 1945463 1904478 > percent 3.0% 10.0% 21.58% 32.3% >=20 > Signed-off-by: Xuan Zhuo > Reviewed-by: Dust Li > --- > drivers/net/virtio_net.c | 2 +- > include/linux/netdev_features.h | 5 +- > net/ethtool/common.c | 1 + > net/xdp/xsk.c | 108 +++++++++++++++++++++++++++++++++-= ------ > 4 files changed, 97 insertions(+), 19 deletions(-) >=20 > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > index 4ecccb8..841a331 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -2985,7 +2985,7 @@ static int virtnet_probe(struct virtio_device *vdev= ) > =09/* Set up network device as normal. */ > =09dev->priv_flags |=3D IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; > =09dev->netdev_ops =3D &virtnet_netdev; > -=09dev->features =3D NETIF_F_HIGHDMA; > +=09dev->features =3D NETIF_F_HIGHDMA | NETIF_F_SKB_NO_LINEAR; > =20 > =09dev->ethtool_ops =3D &virtnet_ethtool_ops; > =09SET_NETDEV_DEV(dev, &vdev->dev); > diff --git a/include/linux/netdev_features.h b/include/linux/netdev_featu= res.h > index 934de56..8dd28e2 100644 > --- a/include/linux/netdev_features.h > +++ b/include/linux/netdev_features.h > @@ -85,9 +85,11 @@ enum { > =20 > =09NETIF_F_HW_MACSEC_BIT,=09=09/* Offload MACsec operations */ > =20 > +=09NETIF_F_SKB_NO_LINEAR_BIT,=09/* Allow skb linear is empty */ > + > =09/* > =09 * Add your fresh new feature above and remember to update > -=09 * netdev_features_strings[] in net/core/ethtool.c and maybe > +=09 * netdev_features_strings[] in net/ethtool/common.c and maybe > =09 * some feature mask #defines below. Please also describe it > =09 * in Documentation/networking/netdev-features.rst. > =09 */ > @@ -157,6 +159,7 @@ enum { > #define NETIF_F_GRO_FRAGLIST=09__NETIF_F(GRO_FRAGLIST) > #define NETIF_F_GSO_FRAGLIST=09__NETIF_F(GSO_FRAGLIST) > #define NETIF_F_HW_MACSEC=09__NETIF_F(HW_MACSEC) > +#define NETIF_F_SKB_NO_LINEAR=09__NETIF_F(SKB_NO_LINEAR) > =20 > /* Finds the next feature with the highest number of the range of start = till 0. > */ > diff --git a/net/ethtool/common.c b/net/ethtool/common.c > index 24036e3..2f3d309 100644 > --- a/net/ethtool/common.c > +++ b/net/ethtool/common.c > @@ -68,6 +68,7 @@ > =09[NETIF_F_HW_TLS_RX_BIT] =3D=09 "tls-hw-rx-offload", > =09[NETIF_F_GRO_FRAGLIST_BIT] =3D=09 "rx-gro-list", > =09[NETIF_F_HW_MACSEC_BIT] =3D=09 "macsec-hw-offload", > +=09[NETIF_F_SKB_NO_LINEAR_BIT] =3D=09 "skb-no-linear", > }; > =20 > const char I think the best would be if you will split this patch into three: - the first one will introduce NETI_F_SKB_NO_LINEAR; - the second will add this feature to virtio_net; - the third will do the rest. Also, it would be nice if you'll mention (in the cover letter or in the third patch) that in order to get a nice boost on non-ZC XSK xmit developers can add a support for completely non-linear skbs and advertise this new feature in their drivers. I think there'll be enough folks wanting to do this. > diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c > index 8037b04..94d17dc 100644 > --- a/net/xdp/xsk.c > +++ b/net/xdp/xsk.c > @@ -430,6 +430,95 @@ 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) > +{ > +=09u32 len, offset, copy, copied; > +=09struct sk_buff *skb; > +=09struct page *page; > +=09char *buffer; > +=09int err, i; > +=09u64 addr; > + > +=09skb =3D sock_alloc_send_skb(&xs->sk, 0, 1, &err); > +=09if (unlikely(!skb)) > +=09=09return NULL; > + > +=09addr =3D desc->addr; > +=09len =3D desc->len; > + > +=09buffer =3D xsk_buff_raw_get_data(xs->pool, addr); > +=09offset =3D offset_in_page(buffer); > +=09addr =3D buffer - (char *)xs->pool->addrs; > + > +=09for (copied =3D 0, i =3D 0; copied < len; ++i) { > +=09=09page =3D xs->pool->umem->pgs[addr >> PAGE_SHIFT]; > + > +=09=09get_page(page); > + > +=09=09copy =3D min((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 len; > + > +=09refcount_add(len, &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, int *err) As the others said, just use ERR_PTR() and PTR_ERR(). You also should have received the letters from kernel test robot that the current version is non-compilable at all. > +=09struct sk_buff *skb; > + > +=09if (xs->dev->features & NETIF_F_SKB_NO_LINEAR) { > +=09=09skb =3D xsk_build_skb_zerocopy(xs, desc); > +=09=09if (unlikely(!skb)) { > +=09=09=09*err =3D -ENOMEM; > +=09=09=09return NULL; > +=09=09} > +=09} else { > +=09=09char *buffer; > +=09=09u64 addr; > +=09=09u32 len; > +=09=09int err; > + > +=09=09len =3D desc->len; > +=09=09skb =3D sock_alloc_send_skb(&xs->sk, len, 1, &err); > +=09=09if (unlikely(!skb)) { > +=09=09=09*err =3D -ENOMEM; > +=09=09=09return NULL; > +=09=09} > + > +=09=09skb_put(skb, len); > +=09=09addr =3D desc->addr; > +=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=09*err =3D -EINVAL; > +=09=09=09return NULL; > +=09=09} > +=09} > + > +=09skb->dev =3D xs->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); > @@ -446,43 +535,28 @@ static int xsk_generic_xmit(struct sock *sk) > =09=09goto out; > =20 > =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, len, 1, &err); > +=09=09skb =3D xsk_build_skb(xs, &desc, &err); > =09=09if (unlikely(!skb)) > =09=09=09goto out; > =20 > -=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 > 1.8.3.1