Hi William, All the NICs with flow director cannot benefit from the patch due to the flow matching limitation in the NIC. The offload patch programs the NIC to match on tunnel + header fields (L2 + L3 + L4 + Tunnel header). The XL710 supports filtering on any flexible 16 byte field after the L4 header that used to match the VxLAN header fields such as tenant ID. However the 10G- 82599 NIC supports only 2 Byte flex field which is not enough to match on tunnel headers.
Regards _Sugesh From: William Tu [mailto:u9012...@gmail.com] Sent: Saturday, March 19, 2016 1:50 AM To: Chandran, Sugesh <sugesh.chand...@intel.com> Cc: Ben Pfaff <b...@ovn.org>; dev@openvswitch.org Subject: Re: [ovs-dev] [RFC PATCH] tunneling: Improving vxlan performance using DPDK flow director feature. Hi Sugesh, I saw many Intel NICs have flow director supports, such as Intel 10G 82599. Instead of i40e, do you think other NICs could also benefit from this hardware offload patch? Thanks. Regards, William On Fri, Mar 18, 2016 at 8:51 AM, Chandran, Sugesh <sugesh.chand...@intel.com<mailto:sugesh.chand...@intel.com>> wrote: Hi Ben Thank you for looking into this. We are working on a generic tunneling offloading APIs and more updates will follow. Regards _Sugesh > -----Original Message----- > From: Ben Pfaff [mailto:b...@ovn.org<mailto:b...@ovn.org>] > Sent: Friday, March 18, 2016 2:53 AM > To: Chandran, Sugesh > <sugesh.chand...@intel.com<mailto:sugesh.chand...@intel.com>> > Cc: dev@openvswitch.org<mailto:dev@openvswitch.org> > Subject: Re: [ovs-dev] [RFC PATCH] tunneling: Improving vxlan performance > using DPDK flow director feature. > > This seems really, really specific to the particular NIC. Can you add a > generic > tunnel offload interface to DPDK? What would that look like? > > On Thu, Mar 17, 2016 at 10:43:42PM +0000, Chandran, Sugesh wrote: > > Hi, > > > > This patch proposes an approach that uses Flow director feature on the > Intel Fortville NICs to boost the VxLAN tunneling performance. In our testing > we verified that the VxLAN performance is almost doubled with this patch. > > The solution programs the NIC to report the flow ID along with the VxLAN > packets, and it is matched by OVS in software. There may be corner cases > that needs to addressed in the approach, For eg: There is a possibility of > race > condition where NIC reports flow ID that may match on different flow in > OVS. This happen when a rule is evicted by a new rule with same flowID+ > hash in the OVS software. The packets may hit on wrong new rule in OVS > until the flow get deleted in the hardware too. > > > > It is a hardware specific implementation (Only work with Intel Fortville > NICs) for now, however the proposal works with any programmable > NICs.This RFC proves that the OVS can offer very high speed tunneling > performance using flow programmability in NICs. I am looking for > comments/suggestions on adding this support(such as configuring, enable it > for all the programmable NICs and etc) in OVS userspace datapath for > improving the performance. > > > > Regards > > _Sugesh > > > > > > > -----Original Message----- > > > From: Chandran, Sugesh > > > Sent: Thursday, March 17, 2016 10:00 PM > > > To: dev@openvswitch.org<mailto:dev@openvswitch.org> > > > Cc: Chandran, Sugesh > > > <sugesh.chand...@intel.com<mailto:sugesh.chand...@intel.com>> > > > Subject: [RFC PATCH] tunneling: Improving vxlan performance using > > > DPDK flow director feature. > > > > > > Optimizing vxlan tunneling performance in userspace datapath using > > > flow director feature in Fortville NIC DPDK ports. OVS uses metadata > > > reported by NIC to improve the flow lookup performance on VxLAN > packets. > > > > > > Signed-off-by: Sugesh Chandran > > > <sugesh.chand...@intel.com<mailto:sugesh.chand...@intel.com>> > > > --- > > > lib/automake.mk<http://automake.mk> | 2 + > > > lib/dpdk-i40e-ofld.c | 266 > > > +++++++++++++++++++++++++++++++++++++++++++++++++++ > > > lib/dpdk-i40e-ofld.h | 59 ++++++++++++ > > > lib/dpif-netdev.c | 118 ++++++++++++++++++++++- > > > lib/netdev-dpdk.c | 41 +++++++- > > > 5 files changed, 481 insertions(+), 5 deletions(-) create mode > > > 100644 lib/dpdk-i40e-ofld.c create mode 100644 lib/dpdk-i40e-ofld.h > > > > > > diff --git a/lib/automake.mk<http://automake.mk> > > > b/lib/automake.mk<http://automake.mk> index > > > 27a1669..da48479 > > > 100644 > > > --- a/lib/automake.mk<http://automake.mk> > > > +++ b/lib/automake.mk<http://automake.mk> > > > @@ -366,6 +366,8 @@ endif > > > > > > if DPDK_NETDEV > > > lib_libopenvswitch_la_SOURCES += \ > > > + lib/dpdk-i40e-ofld.c \ > > > + lib/dpdk-i40e-ofld.h \ > > > lib/netdev-dpdk.c \ > > > lib/netdev-dpdk.h > > > endif > > > diff --git a/lib/dpdk-i40e-ofld.c b/lib/dpdk-i40e-ofld.c new file > > > mode 100644 index 0000000..3ea7084 > > > --- /dev/null > > > +++ b/lib/dpdk-i40e-ofld.c > > > @@ -0,0 +1,266 @@ > > > +/* > > > + * Copyright (c) 2016 Intel Corp. > > > + * > > > + * Licensed under the Apache License, Version 2.0 (the "License"); > > > + * you may not use this file except in compliance with the License. > > > + * You may obtain a copy of the License at: > > > + * > > > + * http://www.apache.org/licenses/LICENSE-2.0 > > > + * > > > + * Unless required by applicable law or agreed to in writing, > > > +software > > > + * distributed under the License is distributed on an "AS IS" > > > +BASIS, > > > + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express > or > > > implied. > > > + * See the License for the specific language governing permissions > > > + and > > > + * limitations under the License. > > > + */ > > > + > > > +#include <config.h> > > > + > > > +#include "dpdk-i40e-ofld.h" > > > +#include "errno.h" > > > +#include "ovs-thread.h" > > > +#include "openvswitch/vlog.h" > > > +#include "netdev-provider.h" > > > +#include "rte_pci_dev_ids.h" > > > +#include "rte_ethdev.h" > > > + > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > +VLOG_DEFINE_THIS_MODULE(dpdk_hw_ofld); > > > + > > > +#define VXLAN_DST_PORT 4789 > > > +#define VXLAN_HLEN 50 > > > +#define MAX_FDIR_RULES 8000 > > > + > > > +static uint32_t total_fdir_ids; > > > +static struct ovs_mutex hw_ofld_mutex = OVS_MUTEX_INITIALIZER; > > > + > > > +/* > > > + * Returns '0' if FDIR IDs reaches max limit. Only 8000 entries are > > > + * supported in FVL. > > > + */ > > > +static inline uint32_t > > > +i40e_fdir_entry_cnt_inc(void) > > > +{ > > > + if (total_fdir_ids < MAX_FDIR_RULES) { > > > + ovs_mutex_lock(&hw_ofld_mutex); > > > + total_fdir_ids++; > > > + ovs_mutex_unlock(&hw_ofld_mutex); > > > + return (total_fdir_ids); > > > + } > > > + return 0; > > > +} > > > + > > > +static inline void > > > +i40e_fdir_entry_cnt_decr(void) > > > +{ > > > + ovs_mutex_lock(&hw_ofld_mutex); > > > + total_fdir_ids ? total_fdir_ids-- : 0; > > > + ovs_mutex_unlock(&hw_ofld_mutex); } > > > + > > > +/* > > > + * Release the hardware offloading functionality from the dpdk-port. > > > + */ > > > +int > > > +dpdk_hw_ofld_port_release(struct netdev_dpdk *dpdk_port) { > > > + ovs_mutex_lock(&hw_ofld_mutex); > > > + set_i40e_ofld_flag(dpdk_port, 0); > > > + ovs_mutex_unlock(&hw_ofld_mutex); > > > + return 0; > > > +} > > > + > > > +int > > > +dpdk_eth_dev_hw_ofld_init(struct netdev_dpdk *dev, > > > + int n_rxq, int n_txq, > > > + struct rte_eth_conf > > > +*port_conf) { > > > + int err = 0; > > > + struct rte_eth_dev_info info; > > > + uint16_t vendor_id, device_id; > > > + > > > + rte_eth_dev_info_get(get_dpdk_port_id(dev), &info); > > > + vendor_id = info.pci_dev->id.vendor_id; > > > + device_id = info.pci_dev->id.device_id; > > > + /* Configure vxlan offload only if its FVL NIC */ > > > + if (vendor_id != PCI_VENDOR_ID_INTEL || device_id != > > > + I40E_DEV_ID_SFP_XL710) { > > > + ovs_mutex_lock(&hw_ofld_mutex); > > > + set_i40e_ofld_flag(dev, 0); > > > + ovs_mutex_unlock(&hw_ofld_mutex); > > > + err = rte_eth_dev_configure(get_dpdk_port_id(dev), > > > + n_rxq, n_txq, port_conf); > > > + return err; > > > + } > > > + ovs_mutex_lock(&hw_ofld_mutex); > > > + set_i40e_ofld_flag(dev, 1); > > > + ovs_mutex_unlock(&hw_ofld_mutex); > > > + /* Configure FVL FDIR VxLAN tunnel handing */ > > > + port_conf->fdir_conf.mode = RTE_FDIR_MODE_PERFECT; > > > + port_conf->fdir_conf.flex_conf.nb_payloads = 1; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].type = > > > RTE_ETH_L4_PAYLOAD; > > > + /* Need to initilize all the 16 flex bytes,no matter; > > > + * what we really using, possibly a DPDK bug?? */ > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[0] = 0; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[1] = 1; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[2] = 2; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[3] = 3; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[4] = 4; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[5] = 5; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[6] = 6; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[7] = 7; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[8] = 8; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[9] = 9; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[10] = 10; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[11] = 11; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[12] = 12; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[13] = 13; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[14] = 14; > > > + port_conf->fdir_conf.flex_conf.flex_set[0].src_offset[15] = 15; > > > + err = rte_eth_dev_configure(get_dpdk_port_id(dev), > > > + n_rxq, n_txq, port_conf); > > > + if (err) { > > > + VLOG_ERR("Failed to configure DPDK port with hardware offload"); > > > + return err; > > > + } > > > + /*Clean all FDIR entries if any */ > > > + err = rte_eth_dev_filter_ctrl(get_dpdk_port_id(dev), > > > + RTE_ETH_FILTER_FDIR, RTE_ETH_FILTER_FLUSH, NULL); > > > + return err; > > > +} > > > + > > > +/* > > > + * Install rules for VxLAN packets in hardware */ int > > > +set_up_hw_offload_port_rule(struct netdev *netdev__, > > > + const struct flow *flow, > > > + const uint32_t hw_flow_id, > > > + const bool is_add_rule) { > > > + int err = 0; > > > + uint8_t flexbytes[RTE_ETH_FDIR_MAX_FLEXLEN] = { 0 }; > > > + uint32_t *vni; > > > + enum rte_filter_op filter_op; > > > + struct rte_eth_fdir_filter entry = { 0 }; > > > + struct netdev_dpdk *netdev; > > > + > > > + netdev = netdev_dpdk_cast(netdev__); > > > + if (is_i40e_ofld_enable(netdev)) { > > > + entry.soft_id = hw_flow_id; > > > + if (!entry.soft_id) { > > > + VLOG_DBG("Invalid flow ID, Cant install rule in the NIC for " > > > + "hardware offload"); > > > + err = ECANCELED; > > > + return err; > > > + } > > > + /* Install rules in NIC only for VxLAN flows */ > > > + if (ntohs(flow->tp_dst) != VXLAN_DST_PORT) { > > > + return 0; > > > + } > > > + entry.action.flex_off = 0; /* use 0 by default */ > > > + entry.input.flow_ext.vlan_tci = 0; //! ignored by i40e fdir > > > + entry.action.behavior = RTE_ETH_FDIR_PASSTHRU; > > > + entry.action.report_status = RTE_ETH_FDIR_REPORT_ID_FLEX_4; > > > + entry.input.flow_type = RTE_ETH_FLOW_NONFRAG_IPV4_UDP; > > > + entry.input.flow.ip4_flow.src_ip = flow->nw_src; > > > + entry.input.flow.ip4_flow.dst_ip = flow->nw_dst; > > > + entry.input.flow.udp4_flow.dst_port = htons(VXLAN_DST_PORT); > > > + entry.input.flow.udp4_flow.src_port = flow->tp_src; > > > + vni = (uint32_t *)&flexbytes[4]; > > > + *vni = flow->tunnel.tun_id << 8; > > > + memcpy(entry.input.flow_ext.flexbytes, flexbytes, > > > + RTE_ETH_FDIR_MAX_FLEXLEN); > > > + entry.action.rx_queue = 0; > > > + filter_op = is_add_rule ? RTE_ETH_FILTER_ADD : > > > + RTE_ETH_FILTER_DELETE; > > > + err = rte_eth_dev_filter_ctrl(get_dpdk_port_id(netdev), > > > + RTE_ETH_FILTER_FDIR, filter_op, &entry); > > > + > > > + /* > > > + * XXX : Delayed the max limit check for flow director entries > > > after > > > + * the configuration. Anyway the rte_eth_dev_filter_ctrl will > > > fail if > > > + * max limit reaches. This can be used for tracking. > > > + */ > > > + if (is_add_rule) { > > > + if (!i40e_fdir_entry_cnt_inc()) { > > > + VLOG_DBG("Cant configure rule on NIC, Flow director " > > > + "entries hits max limit"); > > > + } > > > + } > > > + else { > > > + i40e_fdir_entry_cnt_decr(); > > > + } > > > + if (err < 0) { > > > + VLOG_DBG("flow director programming error in NIC: (%d)\n", > err); > > > + return err; > > > + } > > > + } > > > + return err; > > > +} > > > + > > > +static int > > > +i40e_dpdk_port_get_hw_ofld_pkts(struct > > > + dp_netdev_pmd_thread *pmd, struct dp_packet > > > + **in_packets, struct dp_packet **hw_packets, > > > + struct dp_packet **non_hw_packets, > > > + uint32_t cnt) > > > +{ > > > + int i, hw_pkt_cnt = 0, norm_pkt_cnt = 0; > > > + const struct dp_netdev_flow *flow; > > > + struct rte_mbuf *mbuf; > > > + > > > + for (i = 0; i < cnt; i++) { > > > + mbuf = (struct rte_mbuf *)in_packets[i]; > > > + if (mbuf->ol_flags & PKT_RX_FDIR_ID) { > > > + flow = lookup_hw_offload_flow_for_fdirid(pmd, mbuf, > > > + mbuf->hash.fdir.hi); > > > + if (!flow) { > > > + /* Bogus flow in hw, cannot find it in OVS EMC */ > > > + mbuf->ol_flags &= ~PKT_RX_FDIR_ID; > > > + non_hw_packets[norm_pkt_cnt++] = in_packets[i]; > > > + continue; > > > + } > > > + dp_packet_reset_packet(in_packets[i], VXLAN_HLEN); > > > + mbuf->ol_flags |= PKT_RX_RSS_HASH; > > > + mbuf->hash.rss = hash_finish(mbuf->hash.rss, 1); > > > + hw_packets[hw_pkt_cnt++] = in_packets[i]; > > > + } > > > + else { > > > + non_hw_packets[norm_pkt_cnt++] = in_packets[i]; > > > + } > > > + } > > > + return hw_pkt_cnt; > > > +} > > > + > > > +/* > > > + * Process the packets based on hardware offload configuration */ > > > +void hw_ofld_dp_netdev_input(struct dp_netdev_pmd_thread *pmd, > > > + struct netdev_rxq *netdev_rxq, > > > + struct dp_packet **packets, int cnt, > > > + odp_port_t port_no) { > > > + int hw_pkt_cnt; > > > + struct dp_packet *hw_ofld_packets[NETDEV_MAX_BURST] = { 0 }; > > > + struct dp_packet *orig_packets[NETDEV_MAX_BURST] = { 0 }; > > > + struct netdev_dpdk *netdev = > > > +netdev_dpdk_cast(netdev_rxq->netdev); > > > + > > > + if (is_i40e_ofld_enable(netdev)) { > > > + hw_pkt_cnt = i40e_dpdk_port_get_hw_ofld_pkts(pmd, packets, > > > + > > > hw_ofld_packets, > > > + orig_packets, > > > cnt); > > > + /* Process packet streams separately. */ > > > + if (hw_pkt_cnt) { > > > + dp_netdev_input(pmd, hw_ofld_packets, hw_pkt_cnt, port_no); > > > + } > > > + if (cnt - hw_pkt_cnt) { > > > + dp_netdev_input(pmd, orig_packets, (cnt - hw_pkt_cnt), > port_no); > > > + } > > > + } > > > + else { > > > + dp_netdev_input(pmd, packets, cnt, port_no); > > > + } > > > +} > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > diff --git a/lib/dpdk-i40e-ofld.h b/lib/dpdk-i40e-ofld.h new file > > > mode 100644 index 0000000..1aad246 > > > --- /dev/null > > > +++ b/lib/dpdk-i40e-ofld.h > > > @@ -0,0 +1,59 @@ > > > +/* > > > + * Copyright (c) 2016 Intel Corp. > > > + * > > > + * Licensed under the Apache License, Version 2.0 (the "License"); > > > + * you may not use this file except in compliance with the License. > > > + * You may obtain a copy of the License at: > > > + * > > > + * http://www.apache.org/licenses/LICENSE-2.0 > > > + * > > > + * Unless required by applicable law or agreed to in writing, > > > +software > > > + * distributed under the License is distributed on an "AS IS" > > > +BASIS, > > > + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express > or > > > implied. > > > + * See the License for the specific language governing permissions > > > + and > > > + * limitations under the License. > > > + */ > > > + > > > +#ifndef DPDK_I40E_OFLD_H_ > > > +#define DPDK_I40E_OFLD_H_ > > > + > > > +#include <config.h> > > > + > > > +#include "dp-packet.h" > > > +#include "netdev.h" > > > + > > > +/* > > > + * Macro to enable/disable HW OFFLOAD feature for DPDK. > > > + * 1 :- Enable HW_OFFLOAD support in OVS > > > + * 0 :- Disable HW_OFFLOAD support in OVS */ > > > +#define DPDK_I40E_TNL_OFFLOAD_ENABLE 1 > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + > > > +struct netdev_dpdk; > > > +struct dp_netdev_pmd_thread; > > > +struct dp_netdev_flow; > > > + > > > +struct netdev_dpdk *netdev_dpdk_cast(const struct netdev *netdev); > > > +extern inline bool is_i40e_ofld_enable(const struct netdev_dpdk > > > +*netdev); extern inline void set_i40e_ofld_flag(struct netdev_dpdk > > > +*netdev, bool flag); extern inline int get_dpdk_port_id(struct > > > +netdev_dpdk *dpdk_port); int dpdk_eth_dev_hw_ofld_init(struct > > > netdev_dpdk *dev, int n_rxq, int n_txq, > > > + struct rte_eth_conf *port_conf); int > > > +dpdk_hw_ofld_port_release(struct netdev_dpdk *dpdk_port); int > > > +set_up_hw_offload_port_rule(struct netdev *netdev__, > > > + const struct flow *flow, > > > + const uint32_t hw_flow_id, > > > + const bool is_add_rule); void > > > +hw_ofld_dp_netdev_input(struct dp_netdev_pmd_thread *pmd, > > > + struct netdev_rxq *netdev_rxq, > > > + struct dp_packet **packets, int cnt, > > > + odp_port_t port_no); const struct > > > +dp_netdev_flow *lookup_hw_offload_flow_for_fdirid( > > > + const struct dp_netdev_pmd_thread *pmd, > > > + struct rte_mbuf *mbuf, uint32_t > > > +flow_id); void dp_netdev_input(struct dp_netdev_pmd_thread *, > > > +struct dp_packet > > > **, > > > + int cnt, odp_port_t port_no); > > > + > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE #endif /* > DPDK_I40E_OFLD_H_ > > > +*/ > > > diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c index > > > cf574ad..d79b239 > > > 100644 > > > --- a/lib/dpif-netdev.c > > > +++ b/lib/dpif-netdev.c > > > @@ -70,6 +70,7 @@ > > > #include "util.h" > > > > > > #include "openvswitch/vlog.h" > > > +#include "dpdk-i40e-ofld.h" > > > > > > VLOG_DEFINE_THIS_MODULE(dpif_netdev); > > > > > > @@ -478,7 +479,7 @@ static void dp_netdev_execute_actions(struct > > > dp_netdev_pmd_thread *pmd, > > > bool may_steal, > > > const struct nlattr *actions, > > > size_t actions_len); -static > > > void dp_netdev_input(struct dp_netdev_pmd_thread *, > > > +void dp_netdev_input(struct dp_netdev_pmd_thread *, > > > struct dp_packet **, int cnt, > > > odp_port_t port_no); static void dp_netdev_recirculate(struct > dp_netdev_pmd_thread *, > > > struct dp_packet **, int cnt); @@ > > > -1455,6 +1456,28 @@ dp_netdev_pmd_remove_flow(struct > dp_netdev_pmd_thread *pmd, > > > flow->dead = true; > > > > > > dp_netdev_flow_unref(flow); > > > + > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + struct dp_netdev_port *dp_port; > > > + int err; > > > + odp_port_t in_port = flow->flow.in_port.odp_port; > > > + err = get_port_by_number(pmd->dp, in_port, &dp_port); > > > + if (err) { > > > + VLOG_WARN("Cannot get the port information, hardware > > > +offload may > > > " > > > + "not be functional"); > > > + return; > > > + } > > > + if(strcmp(dp_port->type, "dpdk")) { > > > + /* No hardware offload on a non-DPDK port") */ > > > + return; > > > + } > > > + /* Remove the hardware offload rule if exists.*/ > > > + if(set_up_hw_offload_port_rule(dp_port->netdev, &flow->flow, > > > + dp_netdev_flow_hash(&(flow->ufid)), 0)) { > > > + VLOG_DBG("Failed to delete the hardware offload rule"); > > > + return; > > > + } > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > } > > > > > > static void > > > @@ -2059,6 +2082,32 @@ dp_netdev_flow_add(struct > > > dp_netdev_pmd_thread *pmd, > > > ds_destroy(&ds); > > > } > > > > > > + /* > > > + * Configure the hardware offload for tunnel while flows are getting > > > + * inserted in OVS. > > > + */ > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + struct dp_netdev_port *dp_port; > > > + int err; > > > + odp_port_t in_port = flow->flow.in_port.odp_port; > > > + err = get_port_by_number(pmd->dp, in_port, &dp_port); > > > + if (err) { > > > + VLOG_WARN("Cannot get the port information, Failed to configure > " > > > + "hardware offload"); > > > + goto out; > > > + } > > > + if (strcmp(dp_port->type, "dpdk")) { > > > + /* No hardware offload on a non-DPDK port */ > > > + goto out; > > > + } > > > + /* install the rule in hw, reduntant might overwrite if it exists*/ > > > + if (set_up_hw_offload_port_rule(dp_port->netdev, &flow->flow, > > > + dp_netdev_flow_hash(&flow->ufid), 1)) { > > > + VLOG_ERR("Failed to install the hardware offload rule"); > > > + goto out; > > > + } > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > +out: > > > return flow; > > > } > > > > > > @@ -2575,7 +2624,19 @@ dp_netdev_process_rxq_port(struct > > > dp_netdev_pmd_thread *pmd, > > > *recirc_depth_get() = 0; > > > > > > cycles_count_start(pmd); > > > + > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + /* Check if the source port is DPDK */ > > > + if (packets[0]->source == DPBUF_DPDK) { > > > + hw_ofld_dp_netdev_input(pmd, rxq, packets, cnt, port- > >port_no); > > > + } > > > + else { > > > + dp_netdev_input(pmd, packets, cnt, port->port_no); > > > + } > > > +#else > > > dp_netdev_input(pmd, packets, cnt, port->port_no); > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + > > > cycles_count_end(pmd, PMD_CYCLES_PROCESSING); > > > } else if (error != EAGAIN && error != EOPNOTSUPP) { > > > static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, > > > 5); @@ - > > > 3321,7 +3382,6 @@ dp_netdev_upcall(struct dp_netdev_pmd_thread > *pmd, > > > struct dp_packet *packet_, > > > flow->tunnel.metadata.present.len = > > > orig_tunnel.metadata.present.len; > > > flow->tunnel.flags |= FLOW_TNL_F_UDPIF; > > > } > > > - > > > return err; > > > } > > > > > > @@ -3430,6 +3490,7 @@ emc_processing(struct > dp_netdev_pmd_thread > > > *pmd, struct dp_packet **packets, > > > struct emc_cache *flow_cache = &pmd->flow_cache; > > > struct netdev_flow_key *key = &keys[0]; > > > size_t i, n_missed = 0, n_dropped = 0; > > > + struct rte_mbuf *mbuf; > > > > > > for (i = 0; i < cnt; i++) { > > > struct dp_netdev_flow *flow; @@ -3454,7 +3515,18 @@ > > > emc_processing(struct dp_netdev_pmd_thread *pmd, struct dp_packet > > > **packets, > > > key->len = 0; /* Not computed yet. */ > > > key->hash = dpif_netdev_packet_get_rss_hash(packet, > > > &key->mf); > > > > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + mbuf = (struct rte_mbuf *)packet; > > > + if (mbuf->ol_flags & PKT_RX_FDIR_ID) { > > > + flow = lookup_hw_offload_flow_for_fdirid(pmd, mbuf, 0); > > > + } > > > + else { > > > + flow = emc_lookup(flow_cache, key); > > > + } > > > +#else > > > flow = emc_lookup(flow_cache, key); > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + > > > if (OVS_LIKELY(flow)) { > > > dp_netdev_queue_batches(packet, flow, &key->mf, batches, > > > n_batches); @@ -3651,7 +3723,7 > > > @@ dp_netdev_input__(struct dp_netdev_pmd_thread *pmd, > > > } > > > } > > > > > > -static void > > > +void > > > dp_netdev_input(struct dp_netdev_pmd_thread *pmd, > > > struct dp_packet **packets, int cnt, > > > odp_port_t port_no) @@ -4290,3 +4362,43 @@ > > > dpcls_lookup(const struct dpcls *cls, const struct netdev_flow_key > > > keys[], > > > } > > > return false; /* Some misses. */ > > > } > > > + > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > +/* > > > + * EMC lookup function on 'flow id' reported by NIC. > > > + */ > > > +const struct dp_netdev_flow * > > > +lookup_hw_offload_flow_for_fdirid(const struct > > > + dp_netdev_pmd_thread *pmd, struct rte_mbuf *mbuf, > > > + uint32_t flow_id) > > > +{ > > > + const struct emc_cache *flow_cache = &pmd->flow_cache; > > > + struct netdev_flow_key key; > > > + struct emc_entry *current_entry; > > > + > > > + key.len = 0; > > > + if (OVS_LIKELY(mbuf->ol_flags & PKT_RX_RSS_HASH)) { > > > + key.hash = mbuf->hash.rss; > > > + } > > > + else { > > > + return NULL; > > > + } > > > + EMC_FOR_EACH_POS_WITH_HASH(flow_cache, current_entry, > > > key.hash) { > > > + if (current_entry->key.hash == key.hash > > > + && emc_entry_alive(current_entry)) { > > > + if (OVS_UNLIKELY(flow_id && dp_netdev_flow_hash( > > > + ¤t_entry->flow->ufid) != > > > + flow_id)) { > > > + /* Hash collision in emc, fallback to software path */ > > > + return NULL; > > > + } > > > + return current_entry->flow; > > > + } > > > + } > > > + /* XXX :: An improved classifier lookup needed here without any > miniflow > > > + * extract to keep it performant.Until then fallback to software > > > based > > > + * packet forwarding on EMC miss. > > > + */ > > > + return NULL; > > > +} > > > +#endif /* DPDK_I40E_TNL_OFFLOAD_ENABLE */ > > > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index > > > f402354..2954f83 > > > 100644 > > > --- a/lib/netdev-dpdk.c > > > +++ b/lib/netdev-dpdk.c > > > @@ -56,6 +56,7 @@ > > > #include "rte_mbuf.h" > > > #include "rte_meter.h" > > > #include "rte_virtio_net.h" > > > +#include "dpdk-i40e-ofld.h" > > > > > > VLOG_DEFINE_THIS_MODULE(dpdk); > > > static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); > > > @@ -112,7 +113,7 @@ static char *vhost_sock_dir = NULL; /* Location > of > > > vhost-user sockets */ > > > */ > > > #define VHOST_ENQ_RETRY_USECS 100 > > > > > > -static const struct rte_eth_conf port_conf = { > > > +static struct rte_eth_conf port_conf = { > > > .rxmode = { > > > .mq_mode = ETH_MQ_RX_RSS, > > > .split_hdr_size = 0, > > > @@ -331,6 +332,9 @@ struct netdev_dpdk { > > > > > > /* Identifier used to distinguish vhost devices from each other */ > > > char vhost_id[PATH_MAX]; > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + bool i40e_ofld_enable; /* hardware/NIC offload flag*/ #endif > > > +//DPDK_I40E_TNL_OFFLOAD_ENABLE > > > > > > /* In dpdk_list. */ > > > struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); @@ -346,6 > > > +350,24 @@ struct netdev_rxq_dpdk { > > > int port_id; > > > }; > > > > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE inline bool > > > +is_i40e_ofld_enable(const struct netdev_dpdk *netdev) { > > > + return netdev->i40e_ofld_enable; } > > > + > > > +inline void set_i40e_ofld_flag(struct netdev_dpdk *netdev, > > > + bool flag) { > > > + netdev->i40e_ofld_enable = flag; } > > > + > > > +inline int get_dpdk_port_id(struct netdev_dpdk *dpdk_port) { > > > + return dpdk_port->port_id; > > > +} > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + > > > static bool dpdk_thread_is_pmd(void); > > > > > > static int netdev_dpdk_construct(struct netdev *); @@ -539,10 > > > +561,21 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int > > > n_rxq, int > > > n_txq) > > > VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, > > > n_txq); > > > } > > > > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + diag = (!dev->i40e_ofld_enable && dev->type == DPDK_DEV_ETH) > ? > > > + dpdk_eth_dev_hw_ofld_init(dev, n_rxq, n_txq, > > > &port_conf) > : > > > + rte_eth_dev_configure(dev->port_id, > > > + n_rxq, n_txq, &port_conf); > > > + if (diag) { > > > + /* rte_dev_configure error */ > > > + break; > > > + } > > > +#else > > > diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, > &port_conf); > > > if (diag) { > > > break; > > > } > > > +#endif //DPDK_I40E_TNL_OFFLOAD_ENABLE > > > > > > for (i = 0; i < n_txq; i++) { > > > diag = rte_eth_tx_queue_setup(dev->port_id, i, > > > NIC_PORT_TX_Q_SIZE, @@ -637,7 +670,7 @@ dpdk_eth_dev_init(struct > > > netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) > > > return 0; > > > } > > > > > > -static struct netdev_dpdk * > > > +struct netdev_dpdk * > > > netdev_dpdk_cast(const struct netdev *netdev) { > > > return CONTAINER_OF(netdev, struct netdev_dpdk, up); @@ -861,6 > > > +894,10 @@ netdev_dpdk_destruct(struct netdev *netdev_) > > > rte_free(dev->tx_q); > > > list_remove(&dev->list_node); > > > dpdk_mp_put(dev->dpdk_mp); > > > + > > > +#ifdef DPDK_I40E_TNL_OFFLOAD_ENABLE > > > + dpdk_hw_ofld_port_release(dev); #endif /* > > > +DPDK_I40E_TNL_OFFLOAD_ENABLE */ > > > ovs_mutex_unlock(&dpdk_mutex); > > > } > > > > > > -- > > > 1.9.1 > > > > _______________________________________________ > > dev mailing list > > dev@openvswitch.org<mailto:dev@openvswitch.org> > > http://openvswitch.org/mailman/listinfo/dev _______________________________________________ dev mailing list dev@openvswitch.org<mailto:dev@openvswitch.org> http://openvswitch.org/mailman/listinfo/dev _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
