Tested with connectx-5

Without patch

10Mpps - > 16 cores used

   PerfTop:   66258 irqs/sec  kernel:99.3%  exact:  0.0% [4000Hz cycles],  (all, 32 CPUs) ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

    10.12%  [kernel]       [k] do_raw_spin_lock
     6.31%  [kernel]       [k] fib_table_lookup
     6.12%  [kernel]       [k] mlx5e_handle_rx_cqe_mpwrq
     4.90%  [kernel]       [k] rt_cache_valid
     3.99%  [kernel]       [k] mlx5e_xmit
     3.03%  [kernel]       [k] ip_rcv
     2.68%  [kernel]       [k] __netif_receive_skb_core
     2.54%  [kernel]       [k] skb_dst_force
     2.41%  [kernel]       [k] ip_finish_output2
     2.21%  [kernel]       [k] __build_skb
     2.03%  [kernel]       [k] __dev_queue_xmit
     1.96%  [kernel]       [k] mlx5e_txwqe_complete
     1.79%  [kernel]       [k] ipt_do_table
     1.78%  [kernel]       [k] inet_gro_receive
     1.69%  [kernel]       [k] ip_forward
     1.66%  [kernel]       [k] udp_v4_early_demux
     1.65%  [kernel]       [k] dst_release
     1.56%  [kernel]       [k] ip_rcv_finish
     1.45%  [kernel]       [k] dev_gro_receive
     1.45%  [kernel]       [k] netif_skb_features
     1.39%  [kernel]       [k] mlx5e_poll_tx_cq
     1.35%  [kernel]       [k] mlx5e_txwqe_build_dsegs
     1.35%  [kernel]       [k] ip_route_input_rcu
     1.15%  [kernel]       [k] dev_hard_start_xmit
     1.12%  [kernel]       [k] napi_gro_receive
     1.07%  [kernel]       [k] netif_receive_skb_internal
     0.98%  [kernel]       [k] sch_direct_xmit
     0.95%  [kernel]       [k] kmem_cache_alloc
     0.89%  [kernel]       [k] read_tsc
     0.88%  [kernel]       [k] mlx5e_build_rx_skb
     0.86%  [kernel]       [k] mlx5_cqwq_get_cqe
     0.82%  [kernel]       [k] page_frag_free
     0.78%  [kernel]       [k] __local_bh_enable_ip
     0.69%  [kernel]       [k] skb_network_protocol
     0.68%  [kernel]       [k] __netif_receive_skb
     0.67%  [kernel]       [k] vlan_dev_hard_start_xmit
     0.65%  [kernel]       [k] mlx5e_poll_rx_cq
     0.65%  [kernel]       [k] validate_xmit_skb
     0.60%  [kernel]       [k] eth_type_trans
     0.60%  [kernel]       [k] deliver_ptype_list_skb
     0.60%  [kernel]       [k] fib_validate_source
     0.55%  [kernel]       [k] eth_header
     0.53%  [kernel]       [k] netdev_pick_tx
     0.53%  [kernel]       [k] __napi_alloc_skb
     0.51%  [kernel]       [k] __udp4_lib_lookup
     0.50%  [kernel]       [k] eth_type_vlan
     0.49%  [kernel]       [k] ip_output
     0.49%  [kernel]       [k] page_frag_alloc
     0.49%  [kernel]       [k] ip_finish_output
     0.48%  [kernel]       [k] neigh_connected_output
     0.45%  [kernel]       [k] nf_hook_slow
     0.44%  [kernel]       [k] udp4_gro_receive
     0.39%  [kernel]       [k] mlx5e_features_check
     0.39%  [kernel]       [k] mlx5e_napi_poll
     0.37%  [kernel]       [k] __jhash_nwords
     0.37%  [kernel]       [k] udp_gro_receive
     0.36%  [kernel]       [k] swiotlb_map_page
     0.33%  [kernel]       [k] mlx5_cqwq_get_wqe
     0.33%  [kernel]       [k] __netdev_pick_tx
     0.29%  [kernel]       [k] ktime_get_with_offset
     0.29%  [kernel]       [k] get_dma_ops
     0.29%  [kernel]       [k] validate_xmit_skb_list
     0.26%  [kernel]       [k] vlan_passthru_hard_header
     0.26%  [kernel]       [k] __udp4_lib_lookup_skb
     0.24%  [kernel]       [k] get_dma_ops
     0.24%  [kernel]       [k] skb_release_data
     0.23%  [kernel]       [k] ip_forward_finish
     0.23%  [kernel]       [k] kmem_cache_free_bulk
     0.23%  [kernel]       [k] timekeeping_get_ns
     0.22%  [kernel]       [k] ip_skb_dst_mtu
     0.21%  [kernel]       [k] compound_head
     0.20%  [kernel]       [k] skb_gro_reset_offset
     0.20%  [kernel]       [k] is_swiotlb_buffer
     0.19%  [kernel]       [k] __net_timestamp.isra.90
     0.19%  [kernel]       [k] dst_metric.constprop.61
     0.18%  [kernel]       [k] skb_orphan_frags.constprop.126
     0.18%  [kernel]       [k] _kfree_skb_defer
     0.18%  [kernel]       [k] irq_entries_start
     0.17%  [kernel]       [k] dev_hard_header.constprop.54
     0.17%  [kernel]       [k] dma_mapping_error
     0.17%  [kernel]       [k] neigh_resolve_output




With patch

12Mpps -> 16 cores

   PerfTop:   66209 irqs/sec  kernel:99.3%  exact:  0.0% [4000Hz cycles],  (all, 32 CPUs)
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

    10.67%  [kernel]       [k] do_raw_spin_lock
     6.96%  [kernel]       [k] fib_table_lookup
     6.53%  [kernel]       [k] mlx5e_handle_rx_cqe_mpwrq
     4.17%  [kernel]       [k] mlx5e_xmit
     3.22%  [kernel]       [k] ip_rcv
     3.07%  [kernel]       [k] __netif_receive_skb_core
     2.86%  [kernel]       [k] __dev_queue_xmit
     2.36%  [kernel]       [k] __build_skb
     2.33%  [kernel]       [k] ip_forward
     2.05%  [kernel]       [k] mlx5e_txwqe_complete
     2.02%  [kernel]       [k] ip_finish_output2
     2.00%  [kernel]       [k] ipt_do_table
     1.84%  [kernel]       [k] ip_rcv_finish
     1.83%  [kernel]       [k] inet_gro_receive
     1.80%  [kernel]       [k] udp_v4_early_demux
     1.61%  [kernel]       [k] dev_gro_receive
     1.55%  [kernel]       [k] netif_skb_features
     1.52%  [kernel]       [k] mlx5e_txwqe_build_dsegs
     1.47%  [kernel]       [k] mlx5e_poll_tx_cq
     1.39%  [kernel]       [k] ip_route_input_rcu
     1.38%  [kernel]       [k] dev_hard_start_xmit
     1.17%  [kernel]       [k] netif_receive_skb_internal
     1.16%  [kernel]       [k] napi_gro_receive
     1.03%  [kernel]       [k] kmem_cache_alloc
     1.02%  [kernel]       [k] sch_direct_xmit
     0.97%  [kernel]       [k] read_tsc
     0.94%  [kernel]       [k] page_frag_free
     0.91%  [kernel]       [k] mlx5_cqwq_get_cqe
     0.90%  [kernel]       [k] mlx5e_build_rx_skb
     0.89%  [kernel]       [k] skb_network_protocol
     0.83%  [kernel]       [k] __local_bh_enable_ip
     0.79%  [kernel]       [k] validate_xmit_skb
     0.77%  [kernel]       [k] vlan_dev_hard_start_xmit
     0.74%  [kernel]       [k] __netif_receive_skb
     0.72%  [kernel]       [k] mlx5e_poll_rx_cq
     0.70%  [kernel]       [k] netdev_pick_tx
     0.69%  [kernel]       [k] eth_type_vlan
     0.68%  [kernel]       [k] __netdev_pick_tx
     0.66%  [kernel]       [k] nf_hook_slow
     0.65%  [kernel]       [k] deliver_ptype_list_skb
     0.62%  [kernel]       [k] fib_validate_source
     0.61%  [kernel]       [k] eth_header
     0.60%  [kernel]       [k] eth_type_trans
     0.59%  [kernel]       [k] __udp4_lib_lookup
     0.58%  [kernel]       [k] __napi_alloc_skb
     0.53%  [kernel]       [k] ip_finish_output
     0.51%  [kernel]       [k] neigh_connected_output
     0.50%  [kernel]       [k] ip_output
     0.50%  [kernel]       [k] rt_cache_valid
     0.44%  [kernel]       [k] udp4_gro_receive
     0.43%  [kernel]       [k] mlx5e_napi_poll
     0.40%  [kernel]       [k] udp_gro_receive
     0.40%  [kernel]       [k] page_frag_alloc
     0.40%  [kernel]       [k] __jhash_nwords
     0.39%  [kernel]       [k] swiotlb_map_page
     0.38%  [kernel]       [k] mlx5_cqwq_get_wqe
     0.36%  [kernel]       [k] mlx5e_features_check
     0.32%  [kernel]       [k] get_dma_ops
     0.31%  [kernel]       [k] ktime_get_with_offset
     0.31%  [kernel]       [k] validate_xmit_skb_list
     0.28%  [kernel]       [k] vlan_passthru_hard_header
     0.28%  [kernel]       [k] get_dma_ops
     0.27%  [kernel]       [k] __udp4_lib_lookup_skb
     0.26%  [kernel]       [k] skb_gro_reset_offset
     0.25%  [kernel]       [k] skb_release_data
     0.25%  [kernel]       [k] timekeeping_get_ns
     0.24%  [kernel]       [k] kmem_cache_free_bulk
     0.24%  [kernel]       [k] ip_forward_finish
     0.23%  [kernel]       [k] compound_head
     0.23%  [kernel]       [k] ip_skb_dst_mtu
     0.22%  [kernel]       [k] __net_timestamp.isra.90
     0.22%  [kernel]       [k] is_swiotlb_buffer
     0.21%  [kernel]       [k] neigh_resolve_output
     0.21%  [kernel]       [k] dst_metric.constprop.61
     0.20%  [kernel]       [k] skb_orphan_frags.constprop.126
     0.20%  [kernel]       [k] irq_entries_start
     0.19%  [kernel]       [k] mlx5e_calc_min_inline
     0.19%  [kernel]       [k] dev_hard_header.constprop.54
     0.19%  [kernel]       [k] _kfree_skb_defer
     0.18%  [kernel]       [k] _raw_spin_lock
     0.18%  [kernel]       [k] ip_route_input_noref





W dniu 2017-09-09 o 11:03, Paweł Staszewski pisze:
Hi


Are there any plans to have this fix normally in kernel ?

Or it is mostly only hack - not longterm fix and need to be different ?


All tests that was done shows that without this patch there is about 20-30% network forwarding performance degradation when using vlan interfaces


Thanks
Paweł



W dniu 2017-08-15 o 03:17, Eric Dumazet pisze:
On Mon, 2017-08-14 at 18:07 -0700, Eric Dumazet wrote:

Or try to hack the IFF_XMIT_DST_RELEASE flag on the vlan netdev.
Something like :

diff --git a/net/8021q/vlan_netlink.c b/net/8021q/vlan_netlink.c
index 5e831de3103e2f7092c7fa15534def403bc62fb4..9472de846d5c0960996261cb2843032847fa4bf7 100644
--- a/net/8021q/vlan_netlink.c
+++ b/net/8021q/vlan_netlink.c
@@ -143,6 +143,7 @@ static int vlan_newlink(struct net *src_net, struct net_device *dev,
      vlan->vlan_proto = proto;
      vlan->vlan_id     = nla_get_u16(data[IFLA_VLAN_ID]);
      vlan->real_dev     = real_dev;
+    dev->priv_flags |= (real_dev->priv_flags & IFF_XMIT_DST_RELEASE);
      vlan->flags     = VLAN_FLAG_REORDER_HDR;
        err = vlan_check_real_dev(real_dev, vlan->vlan_proto, vlan->vlan_id);







Reply via email to