Re: [PATCH bpf-next] xsk: build skb by page

[Magnus Karlsson <magnus.karlsson@gmail.com>]

On Mon, Jan 18, 2021 at 3:47 PM Alexander Lobakin <alobakin@pm.me> wrote:
>
> From: Alexander Lobakin <alobakin@pm.me>
> Date: Mon, 18 Jan 2021 13:00:17 +0000
>
> > From: Yunsheng Lin <linyunsheng@huawei.com>
> > Date: Mon, 18 Jan 2021 20:40:52 +0800
> >
> >> On 2021/1/16 10:44, Xuan Zhuo wrote:
> >>> This patch is used to construct skb based on page to save memory copy
> >>> overhead.
> >>>
> >>> This has one problem:
> >>>
> >>> 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:
> >>>
> >>>     mlx5_core 0000:3b:00.1 eth1: Error cqe on cqn 0x817, ci 0x8, qn 0x1dbb, 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
> >>>
> >>> 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.
> >>>
> >>> Finally, I added a feature NETIF_F_SKB_NO_LINEAR to identify whether the
> >>
> >> Does it make sense to use ETHTOOL_TX_COPYBREAK tunable in ethtool to
> >> configure if the data is copied or not?
> >
> > As far as I can grep, only mlx4 supports this, and it has a different
> > meaning in that driver.
> > So I guess a new netdev_feature would be a better solution.
> >
> >>> network card supports the header information of the packet in the frag
> >>> and not in the linear space.
> >>>
> >>> ---------------- 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>
> >>> ---
> >>>  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(-)
> >>>
> >>> 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)
> >>>     /* Set up network device as normal. */
> >>>     dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
> >>>     dev->netdev_ops = &virtnet_netdev;
> >>> -   dev->features = NETIF_F_HIGHDMA;
> >>> +   dev->features = NETIF_F_HIGHDMA | NETIF_F_SKB_NO_LINEAR;
> >>>
> >>>     dev->ethtool_ops = &virtnet_ethtool_ops;
> >>>     SET_NETDEV_DEV(dev, &vdev->dev);
> >>> diff --git a/include/linux/netdev_features.h b/include/linux/netdev_features.h
> >>> index 934de56..8dd28e2 100644
> >>> --- a/include/linux/netdev_features.h
> >>> +++ b/include/linux/netdev_features.h
> >>> @@ -85,9 +85,11 @@ enum {
> >>>
> >>>     NETIF_F_HW_MACSEC_BIT,          /* Offload MACsec operations */
> >>>
> >>> +   NETIF_F_SKB_NO_LINEAR_BIT,      /* Allow skb linear is empty */
> >>> +
> >>>     /*
> >>>      * Add your fresh new feature above and remember to update
> >>> -    * netdev_features_strings[] in net/core/ethtool.c and maybe
> >>> +    * netdev_features_strings[] in net/ethtool/common.c and maybe
> >>>      * some feature mask #defines below. Please also describe it
> >>>      * in Documentation/networking/netdev-features.rst.
> >>>      */
> >>> @@ -157,6 +159,7 @@ enum {
> >>>  #define NETIF_F_GRO_FRAGLIST       __NETIF_F(GRO_FRAGLIST)
> >>>  #define NETIF_F_GSO_FRAGLIST       __NETIF_F(GSO_FRAGLIST)
> >>>  #define NETIF_F_HW_MACSEC  __NETIF_F(HW_MACSEC)
> >>> +#define NETIF_F_SKB_NO_LINEAR      __NETIF_F(SKB_NO_LINEAR)
> >>>
> >>>  /* 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 @@
> >>>     [NETIF_F_HW_TLS_RX_BIT] =        "tls-hw-rx-offload",
> >>>     [NETIF_F_GRO_FRAGLIST_BIT] =     "rx-gro-list",
> >>>     [NETIF_F_HW_MACSEC_BIT] =        "macsec-hw-offload",
> >>> +   [NETIF_F_SKB_NO_LINEAR_BIT] =    "skb-no-linear",
> >
> > I completely forgot to add that you'd better to mention in both
> > enumeration/feature and its Ethtool string that the feature applies
> > to Tx path.
> > Smth like:
> >
> > NETIF_F_SKB_TX_NO_LINEAR{,_BIT}, "skb-tx-no-linear"
> > or
> > NETIF_F_TX_SKB_NO_LINEAR{,_BIT}, "tx-skb-no-linear"
> >
> > Otherwise, it may be confusing for users and developers.
>
> OR, I think we may tight the feature with the new approach to build
> skbs by page as it makes no sense for anything else.
> So, if we define something like:
>
> NETIF_F_XSK_TX_GENERIC_ZC{,_BIT}, "xsk-tx-generic-zerocopy",
>
> then user can toggle your new XSK Tx path on/off via Ethtool for
> drivers that will support it (don't forget to add it to hw_features
> for virtio_net then).

Is there ever a reason not to use this path if it is available? If
not, then it would IMO be better if the xsk base code could probe for
this bit and always use this faster path when the feature is supported
by the driver. If it is not supported by the driver, we will just fall
back on the old approach. This way, the user would get better
performance without having to know it needs to toggle some special bit
using Ethtool.

> >>>  };
> >>>
> >>>  const char
> >>> 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)
> >>>     sock_wfree(skb);
> >>>  }
> >>>
> >>> +static struct sk_buff *xsk_build_skb_zerocopy(struct xdp_sock *xs,
> >>> +                                         struct xdp_desc *desc)
> >>> +{
> >>> +   u32 len, offset, copy, copied;
> >>> +   struct sk_buff *skb;
> >>> +   struct page *page;
> >>> +   char *buffer;
> >>> +   int err, i;
> >>> +   u64 addr;
> >>> +
> >>> +   skb = sock_alloc_send_skb(&xs->sk, 0, 1, &err);
> >>> +   if (unlikely(!skb))
> >>> +           return NULL;
> >>> +
> >>> +   addr = desc->addr;
> >>> +   len = desc->len;
> >>> +
> >>> +   buffer = xsk_buff_raw_get_data(xs->pool, addr);
> >>> +   offset = offset_in_page(buffer);
> >>> +   addr = buffer - (char *)xs->pool->addrs;
> >>> +
> >>> +   for (copied = 0, i = 0; copied < len; ++i) {
> >>> +           page = xs->pool->umem->pgs[addr >> PAGE_SHIFT];
> >>> +
> >>> +           get_page(page);
> >>> +
> >>> +           copy = min((u32)(PAGE_SIZE - offset), len - copied);
> >>> +
> >>> +           skb_fill_page_desc(skb, i, page, offset, copy);
> >>> +
> >>> +           copied += copy;
> >>> +           addr += copy;
> >>> +           offset = 0;
> >>> +   }
> >>> +
> >>> +   skb->len += len;
> >>> +   skb->data_len += len;
> >>> +   skb->truesize += len;
> >>> +
> >>> +   refcount_add(len, &xs->sk.sk_wmem_alloc);
> >>> +
> >>> +   return skb;
> >>> +}
> >>> +
> >>> +static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
> >>> +                                struct xdp_desc *desc, int *err)
> >>> +{
> >>> +   struct sk_buff *skb;
> >>> +
> >>> +   if (xs->dev->features & NETIF_F_SKB_NO_LINEAR) {
> >>> +           skb = xsk_build_skb_zerocopy(xs, desc);
> >>> +           if (unlikely(!skb)) {
> >>> +                   *err = -ENOMEM;
> >>> +                   return NULL;
> >>> +           }
> >>> +   } else {
> >>> +           char *buffer;
> >>> +           u64 addr;
> >>> +           u32 len;
> >>> +           int err;
> >>> +
> >>> +           len = desc->len;
> >>> +           skb = sock_alloc_send_skb(&xs->sk, len, 1, &err);
> >>> +           if (unlikely(!skb)) {
> >>> +                   *err = -ENOMEM;
> >>> +                   return NULL;
> >>> +           }
> >>> +
> >>> +           skb_put(skb, len);
> >>> +           addr = desc->addr;
> >>> +           buffer = xsk_buff_raw_get_data(xs->pool, desc->addr);
> >>> +           err = skb_store_bits(skb, 0, buffer, len);
> >>> +
> >>> +           if (unlikely(err)) {
> >>> +                   kfree_skb(skb);
> >>> +                   *err = -EINVAL;
> >>> +                   return NULL;
> >>> +           }
> >>> +   }
> >>> +
> >>> +   skb->dev = xs->dev;
> >>> +   skb->priority = xs->sk.sk_priority;
> >>> +   skb->mark = xs->sk.sk_mark;
> >>> +   skb_shinfo(skb)->destructor_arg = (void *)(long)desc->addr;
> >>> +   skb->destructor = xsk_destruct_skb;
> >>> +
> >>> +   return skb;
> >>> +}
> >>> +
> >>>  static int xsk_generic_xmit(struct sock *sk)
> >>>  {
> >>>     struct xdp_sock *xs = xdp_sk(sk);
> >>> @@ -446,43 +535,28 @@ static int xsk_generic_xmit(struct sock *sk)
> >>>             goto out;
> >>>
> >>>     while (xskq_cons_peek_desc(xs->tx, &desc, xs->pool)) {
> >>> -           char *buffer;
> >>> -           u64 addr;
> >>> -           u32 len;
> >>> -
> >>>             if (max_batch-- == 0) {
> >>>                     err = -EAGAIN;
> >>>                     goto out;
> >>>             }
> >>>
> >>> -           len = desc.len;
> >>> -           skb = sock_alloc_send_skb(sk, len, 1, &err);
> >>> +           skb = xsk_build_skb(xs, &desc, &err);
> >>>             if (unlikely(!skb))
> >>>                     goto out;
> >>>
> >>> -           skb_put(skb, len);
> >>> -           addr = desc.addr;
> >>> -           buffer = xsk_buff_raw_get_data(xs->pool, addr);
> >>> -           err = skb_store_bits(skb, 0, buffer, len);
> >>>             /* This is the backpressure mechanism for the Tx path.
> >>>              * Reserve space in the completion queue and only proceed
> >>>              * if there is space in it. This avoids having to implement
> >>>              * any buffering in the Tx path.
> >>>              */
> >>>             spin_lock_irqsave(&xs->pool->cq_lock, flags);
> >>> -           if (unlikely(err) || xskq_prod_reserve(xs->pool->cq)) {
> >>> +           if (xskq_prod_reserve(xs->pool->cq)) {
> >>>                     spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
> >>>                     kfree_skb(skb);
> >>>                     goto out;
> >>>             }
> >>>             spin_unlock_irqrestore(&xs->pool->cq_lock, flags);
> >>>
> >>> -           skb->dev = xs->dev;
> >>> -           skb->priority = sk->sk_priority;
> >>> -           skb->mark = sk->sk_mark;
> >>> -           skb_shinfo(skb)->destructor_arg = (void *)(long)desc.addr;
> >>> -           skb->destructor = xsk_destruct_skb;
> >>> -
> >>>             err = __dev_direct_xmit(skb, xs->queue_id);
> >>>             if  (err == NETDEV_TX_BUSY) {
> >>>                     /* Tell user-space to retry the send */
> >>>
> >
> > Al
>
> Al
>