Can I ask what kind of packets you are sending? Because your packets shouldn't hit my code. This is to accept flows for tunneled pkts on outer and inner headers.
-----Original Message----- From: Tu, Lijuan Sent: Thursday, December 6, 2018 8:52 PM To: Zhang, Qi Z <qi.z.zh...@intel.com>; Kar, Subarna <subarna....@intel.com>; dev@dpdk.org; Tu, Lijuan <lijuan...@intel.com> Cc: Wu, Jingjing <jingjing...@intel.com>; Gasparakis, Joseph <joseph.gaspara...@intel.com> Subject: RE: [dpdk-dev] [PATCH] net-i40e-Match-on-outer-and-inner-headers-for-tunnel Hi Subarna, This patch cause a big performance drop , ~30% with single core performance test case. > -----Original Message----- > From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Zhang, Qi Z > Sent: Thursday, December 6, 2018 7:45 PM > To: Kar, Subarna <subarna....@intel.com>; dev@dpdk.org > Cc: Wu, Jingjing <jingjing...@intel.com>; Gasparakis, Joseph > <joseph.gaspara...@intel.com> > Subject: Re: [dpdk-dev] [PATCH] > net-i40e-Match-on-outer-and-inner-headers-for-tunnel > > Hi Subarna: > > > -----Original Message----- > > From: Kar, Subarna > > Sent: Wednesday, December 5, 2018 7:13 PM > > To: dev@dpdk.org > > Cc: Zhang, Qi Z <qi.z.zh...@intel.com>; Wu, Jingjing > > <jingjing...@intel.com>; Kar, Subarna <subarna....@intel.com> > > Subject: [PATCH] > > net-i40e-Match-on-outer-and-inner-headers-for-tunnel > > The title is not following DPDK style > it could be > net/i40e: match on outer and inner headers for tunnel. > > Also please add detail commit log for such a large patch. > > Btw, the patch is too large, it's better to separate it into small > ones > > So far what I can suggestion is, at least you can separate into 2 > patches One for new feature in fdir low level and one for bridge rte_flow to > fdir. > > Below are some captures > > > > From: Subarna Kar <subarna....@intel.com> > > > > --- > > drivers/net/i40e/i40e_ethdev.c | 14 +- > > drivers/net/i40e/i40e_ethdev.h | 109 +++++- > > drivers/net/i40e/i40e_fdir.c | 392 +++++++++++++++++++-- > > drivers/net/i40e/i40e_flow.c | 781 > > ++++++++++++++++++++++++++++++++++++++++- > > 4 files changed, 1250 insertions(+), 46 deletions(-) > > > > diff --git a/drivers/net/i40e/i40e_ethdev.c > > b/drivers/net/i40e/i40e_ethdev.c index 7030eb1..0e9f22d 100644 > > --- a/drivers/net/i40e/i40e_ethdev.c > > +++ b/drivers/net/i40e/i40e_ethdev.c > > @@ -1788,6 +1788,7 @@ i40e_dev_configure(struct rte_eth_dev *dev) > > * legacy filter API is deprecated, the following codes should also be > > * removed. > > */ > > + dev->data->dev_conf.fdir_conf.mode = RTE_FDIR_MODE_PERFECT; > > This is not correct, this will prevent vPMD be used for all cases, I > know vPMD should not be used when the new flow be enabled, and below > patch has some issue will break this, but we should fix on that patch anyway. > https://patches.dpdk.org/patch/45697/ > > > diff --git a/drivers/net/i40e/i40e_ethdev.h > > b/drivers/net/i40e/i40e_ethdev.h index 11ecfc3..f7311ce 100644 > > --- a/drivers/net/i40e/i40e_ethdev.h > > +++ b/drivers/net/i40e/i40e_ethdev.h > > @@ -466,6 +466,92 @@ struct i40e_vmdq_info { #define > > I40E_FLEX_WORD_MASK(off) (0x80 >> (off)) > > #define I40E_FDIR_IPv6_TC_OFFSET 20 > > > > static inline int > > +i40e_flow_fdir_fill_outer_header(const struct i40e_fdir_input *fdir_input, > > + unsigned char *raw_pkt) > > +{ > > + struct ipv4_hdr *ip; > > + struct ipv6_hdr *ip6; > > + struct udp_hdr *udp; > > + struct rte_flow_item_gre *gre; > > + struct rte_flow_item_mpls *mpls; > > + struct rte_flow_item_vxlan *vxlan; > > + uint16_t *ether_type; > > + uint8_t len = 2 * sizeof(struct ether_addr); > > + raw_pkt += 2 * sizeof(struct ether_addr); > > + ether_type = (uint16_t *)raw_pkt; > > + raw_pkt += sizeof(uint16_t); > > + len += sizeof(uint16_t); > > + > > + > > + //we don't need ETH, so direct skip to IP > > It's better to use /* xxx */ to follow coding style And same comment > for all other places (except the case that you see nearby comment is > already broken and you just follow it :)) > > > + vxlan = (struct rte_flow_item_vxlan *)raw_pkt; > > + if (fdir_input->flow_ext.iip_type == I40E_FDIR_IPTYPE_IPV4) { > > + vxlan->vni[0] = > > + fdir_input->flow.vxlanudp4_flow.vxlan.vni[0]; > > + vxlan->vni[1] = > > + fdir_input->flow.vxlanudp4_flow.vxlan.vni[1]; > > + vxlan->vni[2] = > > + fdir_input->flow.vxlanudp4_flow.vxlan.vni[2]; > > + } else { > > + vxlan->vni[0] = > > + fdir_input->flow.vxlan6udp4_flow.vxlan.vni[0]; > > + vxlan->vni[1] = > > + fdir_input->flow.vxlan6udp4_flow.vxlan.vni[1]; > > + vxlan->vni[2] = > > + fdir_input->flow.vxlan6udp4_flow.vxlan.vni[2]; > > + } > > + vxlan->flags = VXLAN_DEFAULT_FLAG; > > + len += sizeof(struct rte_flow_item_vxlan); > > + } else { > > + //now put MPLS info for MPLSoUDP and MPLSoGRE > > The intend is not correct, replace spaces with tab > > > + > > +static int > > +i40e_flow_parse_fdir_tunnel_pattern(struct rte_eth_dev *dev, > > + const struct rte_flow_item *pattern, > > + struct rte_flow_error *error, > > + struct i40e_fdir_filter_conf *filter) > > There is couple intend issue in this function, please double check. > > > > > i = 0; > > @@ -4621,9 +5382,17 @@ i40e_flow_validate(struct rte_eth_dev *dev, > > rte_free(items); > > return -rte_errno; > > } > > - if (parse_filter) > > - ret = parse_filter(dev, attr, items, actions, > > + if (parse_filter) { > > + if (parse_filter != TUNNEL_FDIR_FUNCTION && > > + rss == true) { > > s/rss == true/rss > > > + ret = i40e_parse_rss_filter(dev, attr, pattern, > > + actions, &cons_filter, error); > > + return ret; > > + } else { > > + ret = parse_filter(dev, attr, items, actions, > > error, &cons_filter); > > + } > > + } > > flag = true; > > } while ((ret < 0) && (i < RTE_DIM(i40e_supported_patterns))); > > > > -- > > 2.7.4
