| Message ID | 20221212151616.103274-1-rjarry@redhat.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [ovs-dev,v4] netdev-dpdk: add control plane protection support | expand |
| Context | Check | Description |
|---|---|---|
| ovsrobot/apply-robot | warning | apply and check: warning |
| ovsrobot/github-robot-_Build_and_Test | success | github build: passed |
| ovsrobot/intel-ovs-compilation | success | test: success |
Bleep bloop. Greetings Robin Jarry, I am a robot and I have tried out your patch. Thanks for your contribution. I encountered some error that I wasn't expecting. See the details below. checkpatch: WARNING: Line is 82 characters long (recommended limit is 79) #145 FILE: Documentation/topics/dpdk/phy.rst:149: __ https://doc.dpdk.org/guides-21.11/prog_guide/rte_flow.html#device-compatibility WARNING: Line is 80 characters long (recommended limit is 79) #430 FILE: lib/netdev-dpdk.c:5089: dpdk_cp_prot_rss_configure(struct netdev_dpdk *dev, int rss_n_rxq, bool verbose) Lines checked: 669, Warnings: 2, Errors: 0 Please check this out. If you feel there has been an error, please email aconole@redhat.com Thanks, 0-day Robot
Coucou Robin, On Mon, Dec 12, 2022 at 4:16 PM Robin Jarry <rjarry@redhat.com> wrote: > > Some control protocols are used to maintain link status between > forwarding engines (e.g. LACP). When the system is not sized properly, > the PMD threads may not be able to process all incoming traffic from the > configured Rx queues. When a signaling packet of such protocols is > dropped, it can cause link flapping, worsening the situation. > > Use the RTE flow API to redirect these protocols into a dedicated Rx > queue. The assumption is made that the ratio between control protocol > traffic and user data traffic is very low and thus this dedicated Rx > queue will never get full. The RSS redirection table is re-programmed to > only use the other Rx queues. The RSS table size is stored in the > netdev_dpdk structure at port initialization to avoid requesting the > information again when changing the port configuration. > > The additional Rx queue will be assigned a PMD core like any other Rx > queue. Polling that extra queue may introduce increased latency and > a slight performance penalty at the benefit of preventing link flapping. > > This feature must be enabled per port on specific protocols via the > cp-protection option. This option takes a coma-separated list of > protocol names. It is only supported on ethernet ports. > > If the user has already configured multiple Rx queues on the port, an > additional one will be allocated for control plane packets. If the > hardware cannot satisfy the requested number of requested Rx queues, the > last Rx queue will be assigned for control plane. If only one Rx queue > is available, the cp-protection feature will be disabled. If the > hardware does not support the RTE flow matchers/actions, the feature > will be disabled. > > It cannot be enabled when other_config:hw-offload=true as it may > conflict with the offloaded RTE flows. Similarly, if hw-offload is > enabled while some ports already have cp-protection enabled, the RTE > flow offloading will be disabled on these ports. > > Example use: > > ovs-vsctl add-bond br-phy bond0 phy0 phy1 -- \ > set interface phy0 type=dpdk options:dpdk-devargs=0000:ca:00.0 -- \ > set interface phy0 options:cp-protection=lacp -- \ > set interface phy1 type=dpdk options:dpdk-devargs=0000:ca:00.1 -- \ > set interface phy1 options:cp-protection=lacp > > As a starting point, only one protocol is supported: LACP. Other > protocols can be added in the future. NIC compatibility should be > checked. > > To validate that this works as intended, I used a traffic generator to > generate random traffic slightly above the machine capacity at line rate > on a two ports bond interface. OVS is configured to receive traffic on > two VLANs and pop/push them in a br-int bridge based on tags set on > patch ports. > > +----------------------+ > | DUT | > |+--------------------+| > || br-int || default flow, action=NORMAL > || || > || patch10 patch11 || > |+---|-----------|----+| > | | | | > |+---|-----------|----+| > || patch00 patch01 || > || tag:10 tag:20 || > || || > || br-phy || default flow, action=NORMAL > || || > || bond0 || balance-slb, lacp=passive, lacp-time=fast > || phy0 phy1 || > |+------|-----|-------+| > +-------|-----|--------+ > | | > +-------|-----|--------+ > | port0 port1 | balance L3/L4, lacp=active, lacp-time=fast > | lag | mode trunk VLANs 10, 20 > | | > | switch | > | | > | vlan 10 vlan 20 | mode access > | port2 port3 | > +-----|----------|-----+ > | | > +-----|----------|-----+ > | port0 port1 | Random traffic that is properly balanced > | | across the bond ports in both directions. > | traffic generator | > +----------------------+ > > Without cp-protection, the bond0 links are randomly switching to > "defaulted" when one of the LACP packets sent by the switch is dropped > because the RX queues are full and the PMD threads did not process them > fast enough. When that happens, all traffic must go through a single > link which causes above line rate traffic to be dropped. > > When cp-protection is enabled, no LACP packet is dropped and the bond > links remain enabled at all times, maximizing the throughput. > > This feature may be considered as "QoS". However, it does not work by > limiting the rate of traffic explicitly. It only guarantees that some > protocols have a lower chance of being dropped because the PMD cores > cannot keep up with regular traffic. > > The choice of protocols is limited on purpose. This is not meant to be > configurable by users. Some limited configurability could be considered > in the future but it would expose to more potential issues if users are > accidentally redirecting all traffic in the control plane queue. > > Cc: Christophe Fontaine <cfontain@redhat.com> > Cc: Kevin Traynor <ktraynor@redhat.com> > Signed-off-by: Robin Jarry <rjarry@redhat.com> This is not a full review (I did not look too much into the rte_flow / RSS reta stuff), but I did catch some issues. > --- > v3 -> v4: > > * Rebased on master 739bcf2263b3 ("odp-execute: Fix ipv4 missing > clearing of connection tracking fields.") > * Removed the extra reta_size field from struct netdev_dpdk. > rte_eth_dev_info() is now called every time in > dpdk_cp_prot_rss_configure(). > * Fixed RSS reconfiguration for drivers that have redirection tables > that are smaller than RTE_ETH_RETA_GROUP_SIZE (64). > * Added workaround to avoid imbalanced queues when the reta size is too > small and is not a multiple of the number of RSS queues. > * Fixed panic when deleting ports with cp-protection enabled. > * Using cp-protection with an invalid/unsupported protocol now only > prints a warning in the logs and it is ignored. It no longer causes > the port to be destroyed. > > Documentation/topics/dpdk/phy.rst | 55 +++++ > lib/netdev-dpdk.c | 323 +++++++++++++++++++++++++++++- > vswitchd/vswitch.xml | 26 +++ > 3 files changed, 403 insertions(+), 1 deletion(-) This is a new feature, probably worth adding a NEWS update. Plus, should we declare it experimental until we get more feedback? > > diff --git a/Documentation/topics/dpdk/phy.rst b/Documentation/topics/dpdk/phy.rst > index cb2d5bcb7b3f..e4ae69ec2854 100644 > --- a/Documentation/topics/dpdk/phy.rst > +++ b/Documentation/topics/dpdk/phy.rst > @@ -131,6 +131,61 @@ possible with DPDK acceleration. It is possible to configure multiple Rx queues > for ``dpdk`` ports, thus ensuring this is not a bottleneck for performance. For > information on configuring PMD threads, refer to :doc:`pmd`. > > +Control Plane Protection > +------------------------ > + > +Some control protocols are used to maintain link status between forwarding > +engines. In SDN environments, these packets share the same physical network > +than the user data traffic. > + > +When the system is not sized properly, the PMD threads may not be able to > +process all incoming traffic from the configured Rx queues. When a signaling > +packet of such protocols is dropped, it can cause link flapping, worsening the > +situation. > + > +Some physical NICs can be programmed to put these protocols in a dedicated > +hardware Rx queue using the rte_flow__ API. > + > +__ https://doc.dpdk.org/guides-21.11/prog_guide/rte_flow.html#device-compatibility *22.11 now. > + > +The currently supported control plane protocols are: > + > +``lacp`` > + `Link Aggregation Control Protocol`__. Ether type ``0x8809``. > + > + __ https://www.ieee802.org/3/ad/public/mar99/seaman_1_0399.pdf > + > +.. warning:: > + > + This feature is not compatible with all NICs. Refer to vendor documentation > + for more information. > + > +Control plane protection must be enabled on specific protocols per port. The > +``cp-protection`` option requires a coma separated list of protocol names:: > + > + $ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \ > + options:dpdk-devargs=0000:01:00.0 options:cp-protection=lacp > + > +.. note:: > + > + If multiple Rx queues are already configured, regular RSS (Receive Side > + Scaling) queue balancing is done on all but the extra control plane > + protection queue. > + > +.. tip:: > + > + You can check if control plane protection is supported on a port with the > + following command:: > + > + $ ovs-vsctl get interface dpdk-p0 status > + {cp_protection_queue="2", driver_name=..., rss_queues="0-1"} > + > + If the hardware does not support redirecting control plane traffic to > + a dedicated queue, it will be explicit:: > + > + $ ovs-vsctl get interface dpdk-p0 status > + {cp_protection_queue=unsupported, driver_name=..., rss_queues="0-1"} > + > .. _dpdk-phy-flow-control: > > Flow Control > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c > index fff57f78279a..7bbf629e75c7 100644 > --- a/lib/netdev-dpdk.c > +++ b/lib/netdev-dpdk.c > @@ -417,6 +417,11 @@ enum dpdk_hw_ol_features { > NETDEV_TX_SCTP_CHECKSUM_OFFLOAD = 1 << 4, > }; > > +enum dpdk_cp_prot_flags { > + DPDK_CP_PROT_UNSUPPORTED = 1 << 0, > + DPDK_CP_PROT_LACP = 1 << 1, > +}; > + > /* > * In order to avoid confusion in variables names, following naming convention > * should be used, if possible: > @@ -536,6 +541,13 @@ struct netdev_dpdk { > > /* VF configuration. */ > struct eth_addr requested_hwaddr; > + > + /* Requested control plane protection flags, > + * from the enum set 'dpdk_cp_prot_flags' */ > + uint64_t requested_cp_prot_flags; > + uint64_t cp_prot_flags; > + size_t cp_prot_flows_num; > + struct rte_flow **cp_prot_flows; > ); > > PADDED_MEMBERS(CACHE_LINE_SIZE, > @@ -1315,6 +1327,10 @@ common_construct(struct netdev *netdev, dpdk_port_t port_no, > dev->requested_n_txq = NR_QUEUE; > dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; > dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; > + dev->requested_cp_prot_flags = 0; > + dev->cp_prot_flags = 0; > + dev->cp_prot_flows_num = 0; > + dev->cp_prot_flows = NULL; > > /* Initialize the flow control to NULL */ > memset(&dev->fc_conf, 0, sizeof dev->fc_conf); > @@ -1489,6 +1505,8 @@ common_destruct(struct netdev_dpdk *dev) > ovs_mutex_destroy(&dev->mutex); > } > > +static void dpdk_cp_prot_unconfigure(struct netdev_dpdk *dev); > + > static void > netdev_dpdk_destruct(struct netdev *netdev) > { > @@ -1526,6 +1544,9 @@ netdev_dpdk_destruct(struct netdev *netdev) > } > } > > + /* Destroy any rte flows to allow RXQs to be removed. */ > + dpdk_cp_prot_unconfigure(dev); > + > /* Retrieve eth device data before closing it. */ > rte_eth_dev_info_get(dev->port_id, &dev_info); > > @@ -1910,6 +1931,9 @@ dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) > int new_n_rxq; > > new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); > + if (dev->requested_cp_prot_flags) { I'd rather see if cp has a chance to work, by checking both != 0 and not containing DPDK_CP_PROT_UNSUPPORTED. So: if (dev->requested_cp_prot_flags & ~DPDK_CP_PROT_UNSUPPORTED) This way, there should be no need for the "dev->request_n_rxq-=1" later in netdev_dpdk_reconfigure. > + new_n_rxq += 1; > + } > if (new_n_rxq != dev->requested_n_rxq) { > dev->requested_n_rxq = new_n_rxq; > netdev_request_reconfigure(&dev->up); > @@ -1933,6 +1957,50 @@ dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, > } > } > > +static void > +dpdk_cp_prot_set_config(struct netdev *netdev, struct netdev_dpdk *dev, > + const struct smap *args, char **errp) > +{ > + const char *arg = smap_get_def(args, "cp-protection", ""); > + uint64_t flags = 0; > + char buf[256]; > + char *token, *saveptr; > + > + ovs_strzcpy(buf, arg, sizeof(buf)); > + buf[sizeof(buf) - 1] = '\0'; This is control path stuff => no static buffer, please xstrdup() + free. > + > + token = strtok_r(buf, ",", &saveptr); > + while (token) { > + if (strcmp(token, "lacp") == 0) { > + flags |= DPDK_CP_PROT_LACP; > + } else { > + VLOG_WARN_BUF( > + errp, "%s options:cp-protection unknown protocol '%s'", > + netdev_get_name(netdev), token); > + } > + token = strtok_r(NULL, ",", &saveptr); > + } > + > + if (flags && dev->type != DPDK_DEV_ETH) { > + VLOG_WARN_BUF( errp, > + "%s options:cp-protection is only supported on ethernet ports", > + netdev_get_name(netdev)); > + flags = 0; > + } > + > + if (flags && netdev_is_flow_api_enabled()) { > + VLOG_WARN_BUF(errp, > + "%s options:cp-protection is incompatible with hw-offload", > + netdev_get_name(netdev)); > + flags = 0; > + } > + > + if (flags != dev->requested_cp_prot_flags) { > + dev->requested_cp_prot_flags = flags; > + netdev_request_reconfigure(netdev); > + } > +} > + > static int > netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, > char **errp) > @@ -1952,6 +2020,9 @@ netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, > ovs_mutex_lock(&dpdk_mutex); > ovs_mutex_lock(&dev->mutex); > > + /* needs to be called before rxq_config */ > + dpdk_cp_prot_set_config(netdev, dev, args, errp); > + > dpdk_set_rxq_config(dev, args); > > dpdk_process_queue_size(netdev, args, "n_rxq_desc", > @@ -3646,8 +3717,10 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) > struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); > struct rte_eth_dev_info dev_info; > const char *bus_info; > + uint64_t cp_prot_flags; > uint32_t link_speed; > uint32_t dev_flags; > + int n_rxq; > > if (!rte_eth_dev_is_valid_port(dev->port_id)) { > return ENODEV; > @@ -3659,6 +3732,8 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) > link_speed = dev->link.link_speed; > dev_flags = *dev_info.dev_flags; > bus_info = rte_dev_bus_info(dev_info.device); > + cp_prot_flags = dev->cp_prot_flags; > + n_rxq = netdev->n_rxq; > ovs_mutex_unlock(&dev->mutex); > ovs_mutex_unlock(&dpdk_mutex); > > @@ -3701,6 +3776,24 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) > ETH_ADDR_ARGS(dev->hwaddr)); > } > > + if (cp_prot_flags) { > + if (cp_prot_flags & DPDK_CP_PROT_UNSUPPORTED) { > + smap_add(args, "cp_protection_queue", "unsupported"); > + if (n_rxq > 1) { > + smap_add_format(args, "rss_queues", "0-%d", n_rxq - 1); > + } else { > + smap_add(args, "rss_queues", "0"); > + } > + } else { > + smap_add_format(args, "cp_protection_queue", "%d", n_rxq - 1); > + if (n_rxq > 2) { > + smap_add_format(args, "rss_queues", "0-%d", n_rxq - 2); > + } else { > + smap_add(args, "rss_queues", "0"); > + } > + } > + } > + There is something wrong in the status after disabling the feature, so I suspect cp_prot_flags is not correctly cleared when unconfiguring. Before disabling: options : {cp-protection=lacp, dpdk-devargs="0000:82:00.0", n_rxq="2"} status : {bus_info="bus_name=pci, vendor_id=15b3, device_id=1019", cp_protection_queue="2", driver_name=mlx5_pci, if_descr="DPDK 22.11.1 mlx5_pci", if_type="6", link_speed="100Gbps", max_hash_mac_addrs="0", max_mac_addrs="128", max_rx_pktlen="1518", max_rx_queues="1024", max_tx_queues="1024", max_vfs="0", max_vmdq_pools="0", min_rx_bufsize="32", numa_id="1", port_no="0", rss_queues="0-1"} After disabling: options : {dpdk-devargs="0000:82:00.0", n_rxq="2"} status : {bus_info="bus_name=pci, vendor_id=15b3, device_id=1019", cp_protection_queue="1", driver_name=mlx5_pci, if_descr="DPDK 22.11.1 mlx5_pci", if_type="6", link_speed="100Gbps", max_hash_mac_addrs="0", max_mac_addrs="128", max_rx_pktlen="1518", max_rx_queues="1024", max_tx_queues="1024", max_vfs="0", max_vmdq_pools="0", min_rx_bufsize="32", numa_id="1", port_no="0", rss_queues="0"} I still see cp_protection_queue / rss_queues fields in the status colum. > return 0; > } > > @@ -4931,6 +5024,211 @@ static const struct dpdk_qos_ops trtcm_policer_ops = { > .qos_queue_dump_state_init = trtcm_policer_qos_queue_dump_state_init > }; > > +static int > +dpdk_cp_prot_add_flow(struct netdev_dpdk *dev, > + const struct rte_flow_attr *attr, > + const struct rte_flow_item items[], > + const struct rte_flow_action actions[], > + const char *desc, bool dry_run) > +{ > + struct rte_flow_error error; > + struct rte_flow *flow; > + size_t num; > + > + if (dry_run) { > + int ret; > + ret = rte_flow_validate(dev->port_id, attr, items, actions, &error); > + if (rte_flow_validate(dev->port_id, attr, items, actions, &error)) { Hopefully it will yield the same result, but as an optimisation, I propose 'if (ret) {' :-). > + VLOG_WARN("%s: cp-protection: device does not support %s flow: %s", > + netdev_get_name(&dev->up), desc, error.message); > + } > + return ret; > + } > + > + flow = rte_flow_create(dev->port_id, attr, items, actions, &error); > + if (flow == NULL) { > + VLOG_WARN("%s: cp-protection: failed to add %s flow: %s", > + netdev_get_name(&dev->up), desc, error.message); > + return rte_errno; > + } > + > + num = dev->cp_prot_flows_num + 1; > + dev->cp_prot_flows = xrealloc(dev->cp_prot_flows, sizeof(flow) * num); > + dev->cp_prot_flows[dev->cp_prot_flows_num] = flow; > + dev->cp_prot_flows_num = num; > + > + return 0; > +} > + > +static int > +dpdk_cp_prot_add_traffic_flow(struct netdev_dpdk *dev, > + const struct rte_flow_item items[], > + const char *desc, bool dry_run) > +{ > + const struct rte_flow_attr attr = { .ingress = 1 }; > + const struct rte_flow_action actions[] = { > + { > + .type = RTE_FLOW_ACTION_TYPE_QUEUE, > + .conf = &(const struct rte_flow_action_queue) { > + .index = dev->up.n_rxq - 1, > + }, > + }, > + { .type = RTE_FLOW_ACTION_TYPE_END }, > + }; > + > + if (!dry_run) { > + VLOG_INFO("%s: cp-protection: redirecting %s traffic to queue %d", > + netdev_get_name(&dev->up), desc, dev->up.n_rxq - 1); > + } > + return dpdk_cp_prot_add_flow(dev, &attr, items, actions, desc, dry_run); > +} > + > +#define RETA_CONF_SIZE (RTE_ETH_RSS_RETA_SIZE_512 / RTE_ETH_RETA_GROUP_SIZE) > + > +static int > +dpdk_cp_prot_rss_configure(struct netdev_dpdk *dev, int rss_n_rxq, bool verbose) Checkpatch complains about 80 columns limit here. > +{ > + struct rte_eth_rss_reta_entry64 reta_conf[RETA_CONF_SIZE]; > + struct rte_eth_dev_info info; > + int err; > + > + if (rss_n_rxq == 1 && verbose) { > + VLOG_INFO("%s: cp-protection: redirecting other traffic to queue 0", > + netdev_get_name(&dev->up)); > + } else if (verbose) { > + VLOG_INFO("%s: cp-protection: applying rss on queues 0-%d", > + netdev_get_name(&dev->up), rss_n_rxq - 1); > + } > + > + memset(&info, 0, sizeof(info)); Unneeded, dpdk clears this param itself. > + err = rte_eth_dev_info_get(dev->port_id, &info); > + if (err < 0) { > + goto out; > + } > + > + if (info.reta_size % rss_n_rxq != 0 && > + info.reta_size < RTE_ETH_RSS_RETA_SIZE_128) { > + /* > + * Some drivers set reta_size equal to the total number of rxqs that > + * are configured when it is a power of two. Since we are actually > + * reconfiguring the redirection table to exclude the last rxq, we may > + * end up with an imbalanced redirection table. For example, such > + * configuration: > + * > + * options:n_rxq=3 options:cp-protection=lacp > + * > + * Will actually configure 4 rxqs on the NIC, and the default reta to: > + * > + * [0, 1, 2, 3] > + * > + * And dpdk_cp_prot_rss_configure() will reconfigure reta to: > + * > + * [0, 1, 2, 0] > + * > + * Causing queue 0 to receive twice as much traffic as queues 1 and 2. > + * > + * Work around that corner case by forcing a bigger redirection table > + * size to 128 entries when reta_size is not a multiple of rss_n_rxq > + * and when reta_size is less than 128. This value seems to be > + * supported by most of the drivers that also support rte flow. > + */ > + info.reta_size = RTE_ETH_RSS_RETA_SIZE_128; > + } > + > + memset(reta_conf, 0, sizeof(reta_conf)); > + for (uint16_t i = 0; i < info.reta_size; i++) { > + uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE; > + uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE; > + reta_conf[idx].mask |= 1ULL << shift; > + reta_conf[idx].reta[shift] = i % rss_n_rxq; > + } > + err = rte_eth_dev_rss_reta_update(dev->port_id, reta_conf, info.reta_size); > + > +out: > + if (err < 0 && verbose) { > + VLOG_DBG("%s: failed to configure RSS redirection table: err=%d", > + netdev_get_name(&dev->up), err); > + } > + return err; > +} > + > +static int > +dpdk_cp_prot_configure(struct netdev_dpdk *dev, bool dry_run) > +{ > + int err = 0; > + > + if (dev->requested_cp_prot_flags & DPDK_CP_PROT_UNSUPPORTED) { > + goto out; > + } > + if (dev->up.n_rxq < 2) { > + err = ENOTSUP; > + VLOG_DBG("%s: cp-protection: not enough available rx queues", > + netdev_get_name(&dev->up)); > + goto out; > + } > + > + if (dev->requested_cp_prot_flags & DPDK_CP_PROT_LACP) { > + err = dpdk_cp_prot_add_traffic_flow( > + dev, > + (const struct rte_flow_item []) { > + { > + .type = RTE_FLOW_ITEM_TYPE_ETH, > + .spec = &(const struct rte_flow_item_eth){ > + .type = htons(ETH_TYPE_LACP), > + }, > + .mask = &(const struct rte_flow_item_eth){ > + .type = htons(0xffff), > + }, > + }, > + { .type = RTE_FLOW_ITEM_TYPE_END }, > + }, > + "lacp", > + dry_run > + ); > + if (err) { > + goto out; > + } > + } > + > + if (!dry_run && dev->cp_prot_flows_num) { > + /* reconfigure RSS reta in all but the cp protection queue */ > + err = dpdk_cp_prot_rss_configure(dev, dev->up.n_rxq - 1, true); > + } > + > +out: > + if (!dry_run) { > + dev->cp_prot_flags = dev->requested_cp_prot_flags; > + } > + if (err) { > + dev->requested_cp_prot_flags |= DPDK_CP_PROT_UNSUPPORTED; > + } > + return err; > +} > + > +static void > +dpdk_cp_prot_unconfigure(struct netdev_dpdk *dev) > +{ > + struct rte_flow_error error; > + > + if (dev->cp_prot_flows_num == 0) { > + return; > + } > + > + VLOG_DBG("%s: cp-protection: reset flows", netdev_get_name(&dev->up)); > + > + for (int i = 0; i < dev->cp_prot_flows_num; i++) { > + if (rte_flow_destroy(dev->port_id, dev->cp_prot_flows[i], &error)) { > + VLOG_DBG("%s: cp-protection: failed to destroy flow: %s", > + netdev_get_name(&dev->up), error.message); > + } > + } > + free(dev->cp_prot_flows); > + dev->cp_prot_flows_num = 0; > + dev->cp_prot_flows = NULL; > + > + (void) dpdk_cp_prot_rss_configure(dev, dev->up.n_rxq, false); Please log a debug message if it fails. > +} > + > static int > netdev_dpdk_reconfigure(struct netdev *netdev) > { > @@ -4941,6 +5239,7 @@ netdev_dpdk_reconfigure(struct netdev *netdev) > > if (netdev->n_txq == dev->requested_n_txq > && netdev->n_rxq == dev->requested_n_rxq > + && dev->cp_prot_flags == dev->requested_cp_prot_flags > && dev->mtu == dev->requested_mtu > && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode > && dev->rxq_size == dev->requested_rxq_size > @@ -4968,6 +5267,8 @@ netdev_dpdk_reconfigure(struct netdev *netdev) > goto out; > } > > + dpdk_cp_prot_unconfigure(dev); > + > dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; > > netdev->n_txq = dev->requested_n_txq; > @@ -5012,6 +5313,20 @@ netdev_dpdk_reconfigure(struct netdev *netdev) > if (!dev->tx_q) { > err = ENOMEM; > } > + if (!err && dev->requested_cp_prot_flags) { > + /* dry run first */ > + err = dpdk_cp_prot_configure(dev, true); > + if (!err) { > + /* if no error, apply configuration */ > + err = dpdk_cp_prot_configure(dev, false); I don't like this dry_run stuff, it makes the code hard to read... At least move the dev->cp_prot_flags out of this dpdk_cp_prot_configure helper. And instead, store/clear dev->cp_prot_flags only in netdev_dpdk_reconfigure. > + } > + if (err) { > + /* no hw support, remove the extra queue & recover gracefully */ > + err = 0; > + dev->requested_n_rxq -= 1; It should be unneeded if dpdk_set_rxq_config checks for UNSUPPORTED as I proposed earlier. > + netdev_request_reconfigure(netdev); > + } > + } I suspect a dev->cp_prot_flags = 0; is missing here and this is likely the cause for the wrong status once disabling. > > netdev_change_seq_changed(netdev); > > @@ -5213,7 +5528,13 @@ netdev_dpdk_flow_api_supported(struct netdev *netdev) > ovs_mutex_lock(&dev->mutex); > if (dev->type == DPDK_DEV_ETH) { > /* TODO: Check if we able to offload some minimal flow. */ > - ret = true; > + if (dev->requested_cp_prot_flags || dev->cp_prot_flags) { I am unclear if we need to check anything here, but checking requested_cp_prot_flags, only, should be enough. > + VLOG_WARN( > + "%s: hw-offload is mutually exclusive with cp-protection", > + netdev_get_name(netdev)); > + } else { > + ret = true; > + } > } > ovs_mutex_unlock(&dev->mutex); > out: > diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml > index f9bdb2d92bec..e34a1728108f 100644 > --- a/vswitchd/vswitch.xml > +++ b/vswitchd/vswitch.xml > @@ -3430,6 +3430,32 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \ > <p>This option may only be used with dpdk VF representors.</p> > </column> > > + <column name="options" key="cp-protection" > + type='{"type": "string", "enum": ["set", ["lacp"]]}'> > + <p> > + Allocate an extra Rx queue for control plane packets of the specified > + protocol(s). > + </p> > + <p> > + If the user has already configured multiple > + <code>options:n_rxq</code> on the port, an additional one will be > + allocated for control plane packets. If the hardware cannot satisfy > + the requested number of requested Rx queues, the last Rx queue will > + be assigned for control plane. If only one Rx queue is available or > + if the hardware does not support the RTE flow matchers/actions > + required to redirect the selected protocols, > + <code>cp-protection</code> will be disabled. > + </p> > + <p> > + This feature is multually exclusive with > + <code>other_options:hw-offload</code> as it may conflict with the > + offloaded RTE flows. > + </p> > + <p> > + Disabled by default. > + </p> > + </column> > + > <column name="other_config" key="tx-steering" > type='{"type": "string", > "enum": ["set", ["thread", "hash"]]}'> > -- > 2.38.1 > > _______________________________________________ > dev mailing list > dev@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev >
diff --git a/Documentation/topics/dpdk/phy.rst b/Documentation/topics/dpdk/phy.rst index cb2d5bcb7b3f..e4ae69ec2854 100644 --- a/Documentation/topics/dpdk/phy.rst +++ b/Documentation/topics/dpdk/phy.rst @@ -131,6 +131,61 @@ possible with DPDK acceleration. It is possible to configure multiple Rx queues for ``dpdk`` ports, thus ensuring this is not a bottleneck for performance. For information on configuring PMD threads, refer to :doc:`pmd`. +Control Plane Protection +------------------------ + +Some control protocols are used to maintain link status between forwarding +engines. In SDN environments, these packets share the same physical network +than the user data traffic. + +When the system is not sized properly, the PMD threads may not be able to +process all incoming traffic from the configured Rx queues. When a signaling +packet of such protocols is dropped, it can cause link flapping, worsening the +situation. + +Some physical NICs can be programmed to put these protocols in a dedicated +hardware Rx queue using the rte_flow__ API. + +__ https://doc.dpdk.org/guides-21.11/prog_guide/rte_flow.html#device-compatibility + +The currently supported control plane protocols are: + +``lacp`` + `Link Aggregation Control Protocol`__. Ether type ``0x8809``. + + __ https://www.ieee802.org/3/ad/public/mar99/seaman_1_0399.pdf + +.. warning:: + + This feature is not compatible with all NICs. Refer to vendor documentation + for more information. + +Control plane protection must be enabled on specific protocols per port. The +``cp-protection`` option requires a coma separated list of protocol names:: + + $ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \ + options:dpdk-devargs=0000:01:00.0 options:cp-protection=lacp + +.. note:: + + If multiple Rx queues are already configured, regular RSS (Receive Side + Scaling) queue balancing is done on all but the extra control plane + protection queue. + +.. tip:: + + You can check if control plane protection is supported on a port with the + following command:: + + $ ovs-vsctl get interface dpdk-p0 status + {cp_protection_queue="2", driver_name=..., rss_queues="0-1"} + + If the hardware does not support redirecting control plane traffic to + a dedicated queue, it will be explicit:: + + $ ovs-vsctl get interface dpdk-p0 status + {cp_protection_queue=unsupported, driver_name=..., rss_queues="0-1"} + .. _dpdk-phy-flow-control: Flow Control diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index fff57f78279a..7bbf629e75c7 100644 --- a/lib/netdev-dpdk.c +++ b/lib/netdev-dpdk.c @@ -417,6 +417,11 @@ enum dpdk_hw_ol_features { NETDEV_TX_SCTP_CHECKSUM_OFFLOAD = 1 << 4, }; +enum dpdk_cp_prot_flags { + DPDK_CP_PROT_UNSUPPORTED = 1 << 0, + DPDK_CP_PROT_LACP = 1 << 1, +}; + /* * In order to avoid confusion in variables names, following naming convention * should be used, if possible: @@ -536,6 +541,13 @@ struct netdev_dpdk { /* VF configuration. */ struct eth_addr requested_hwaddr; + + /* Requested control plane protection flags, + * from the enum set 'dpdk_cp_prot_flags' */ + uint64_t requested_cp_prot_flags; + uint64_t cp_prot_flags; + size_t cp_prot_flows_num; + struct rte_flow **cp_prot_flows; ); PADDED_MEMBERS(CACHE_LINE_SIZE, @@ -1315,6 +1327,10 @@ common_construct(struct netdev *netdev, dpdk_port_t port_no, dev->requested_n_txq = NR_QUEUE; dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; + dev->requested_cp_prot_flags = 0; + dev->cp_prot_flags = 0; + dev->cp_prot_flows_num = 0; + dev->cp_prot_flows = NULL; /* Initialize the flow control to NULL */ memset(&dev->fc_conf, 0, sizeof dev->fc_conf); @@ -1489,6 +1505,8 @@ common_destruct(struct netdev_dpdk *dev) ovs_mutex_destroy(&dev->mutex); } +static void dpdk_cp_prot_unconfigure(struct netdev_dpdk *dev); + static void netdev_dpdk_destruct(struct netdev *netdev) { @@ -1526,6 +1544,9 @@ netdev_dpdk_destruct(struct netdev *netdev) } } + /* Destroy any rte flows to allow RXQs to be removed. */ + dpdk_cp_prot_unconfigure(dev); + /* Retrieve eth device data before closing it. */ rte_eth_dev_info_get(dev->port_id, &dev_info); @@ -1910,6 +1931,9 @@ dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) int new_n_rxq; new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); + if (dev->requested_cp_prot_flags) { + new_n_rxq += 1; + } if (new_n_rxq != dev->requested_n_rxq) { dev->requested_n_rxq = new_n_rxq; netdev_request_reconfigure(&dev->up); @@ -1933,6 +1957,50 @@ dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, } } +static void +dpdk_cp_prot_set_config(struct netdev *netdev, struct netdev_dpdk *dev, + const struct smap *args, char **errp) +{ + const char *arg = smap_get_def(args, "cp-protection", ""); + uint64_t flags = 0; + char buf[256]; + char *token, *saveptr; + + ovs_strzcpy(buf, arg, sizeof(buf)); + buf[sizeof(buf) - 1] = '\0'; + + token = strtok_r(buf, ",", &saveptr); + while (token) { + if (strcmp(token, "lacp") == 0) { + flags |= DPDK_CP_PROT_LACP; + } else { + VLOG_WARN_BUF( + errp, "%s options:cp-protection unknown protocol '%s'", + netdev_get_name(netdev), token); + } + token = strtok_r(NULL, ",", &saveptr); + } + + if (flags && dev->type != DPDK_DEV_ETH) { + VLOG_WARN_BUF( errp, + "%s options:cp-protection is only supported on ethernet ports", + netdev_get_name(netdev)); + flags = 0; + } + + if (flags && netdev_is_flow_api_enabled()) { + VLOG_WARN_BUF(errp, + "%s options:cp-protection is incompatible with hw-offload", + netdev_get_name(netdev)); + flags = 0; + } + + if (flags != dev->requested_cp_prot_flags) { + dev->requested_cp_prot_flags = flags; + netdev_request_reconfigure(netdev); + } +} + static int netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, char **errp) @@ -1952,6 +2020,9 @@ netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, ovs_mutex_lock(&dpdk_mutex); ovs_mutex_lock(&dev->mutex); + /* needs to be called before rxq_config */ + dpdk_cp_prot_set_config(netdev, dev, args, errp); + dpdk_set_rxq_config(dev, args); dpdk_process_queue_size(netdev, args, "n_rxq_desc", @@ -3646,8 +3717,10 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); struct rte_eth_dev_info dev_info; const char *bus_info; + uint64_t cp_prot_flags; uint32_t link_speed; uint32_t dev_flags; + int n_rxq; if (!rte_eth_dev_is_valid_port(dev->port_id)) { return ENODEV; @@ -3659,6 +3732,8 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) link_speed = dev->link.link_speed; dev_flags = *dev_info.dev_flags; bus_info = rte_dev_bus_info(dev_info.device); + cp_prot_flags = dev->cp_prot_flags; + n_rxq = netdev->n_rxq; ovs_mutex_unlock(&dev->mutex); ovs_mutex_unlock(&dpdk_mutex); @@ -3701,6 +3776,24 @@ netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) ETH_ADDR_ARGS(dev->hwaddr)); } + if (cp_prot_flags) { + if (cp_prot_flags & DPDK_CP_PROT_UNSUPPORTED) { + smap_add(args, "cp_protection_queue", "unsupported"); + if (n_rxq > 1) { + smap_add_format(args, "rss_queues", "0-%d", n_rxq - 1); + } else { + smap_add(args, "rss_queues", "0"); + } + } else { + smap_add_format(args, "cp_protection_queue", "%d", n_rxq - 1); + if (n_rxq > 2) { + smap_add_format(args, "rss_queues", "0-%d", n_rxq - 2); + } else { + smap_add(args, "rss_queues", "0"); + } + } + } + return 0; } @@ -4931,6 +5024,211 @@ static const struct dpdk_qos_ops trtcm_policer_ops = { .qos_queue_dump_state_init = trtcm_policer_qos_queue_dump_state_init }; +static int +dpdk_cp_prot_add_flow(struct netdev_dpdk *dev, + const struct rte_flow_attr *attr, + const struct rte_flow_item items[], + const struct rte_flow_action actions[], + const char *desc, bool dry_run) +{ + struct rte_flow_error error; + struct rte_flow *flow; + size_t num; + + if (dry_run) { + int ret; + ret = rte_flow_validate(dev->port_id, attr, items, actions, &error); + if (rte_flow_validate(dev->port_id, attr, items, actions, &error)) { + VLOG_WARN("%s: cp-protection: device does not support %s flow: %s", + netdev_get_name(&dev->up), desc, error.message); + } + return ret; + } + + flow = rte_flow_create(dev->port_id, attr, items, actions, &error); + if (flow == NULL) { + VLOG_WARN("%s: cp-protection: failed to add %s flow: %s", + netdev_get_name(&dev->up), desc, error.message); + return rte_errno; + } + + num = dev->cp_prot_flows_num + 1; + dev->cp_prot_flows = xrealloc(dev->cp_prot_flows, sizeof(flow) * num); + dev->cp_prot_flows[dev->cp_prot_flows_num] = flow; + dev->cp_prot_flows_num = num; + + return 0; +} + +static int +dpdk_cp_prot_add_traffic_flow(struct netdev_dpdk *dev, + const struct rte_flow_item items[], + const char *desc, bool dry_run) +{ + const struct rte_flow_attr attr = { .ingress = 1 }; + const struct rte_flow_action actions[] = { + { + .type = RTE_FLOW_ACTION_TYPE_QUEUE, + .conf = &(const struct rte_flow_action_queue) { + .index = dev->up.n_rxq - 1, + }, + }, + { .type = RTE_FLOW_ACTION_TYPE_END }, + }; + + if (!dry_run) { + VLOG_INFO("%s: cp-protection: redirecting %s traffic to queue %d", + netdev_get_name(&dev->up), desc, dev->up.n_rxq - 1); + } + return dpdk_cp_prot_add_flow(dev, &attr, items, actions, desc, dry_run); +} + +#define RETA_CONF_SIZE (RTE_ETH_RSS_RETA_SIZE_512 / RTE_ETH_RETA_GROUP_SIZE) + +static int +dpdk_cp_prot_rss_configure(struct netdev_dpdk *dev, int rss_n_rxq, bool verbose) +{ + struct rte_eth_rss_reta_entry64 reta_conf[RETA_CONF_SIZE]; + struct rte_eth_dev_info info; + int err; + + if (rss_n_rxq == 1 && verbose) { + VLOG_INFO("%s: cp-protection: redirecting other traffic to queue 0", + netdev_get_name(&dev->up)); + } else if (verbose) { + VLOG_INFO("%s: cp-protection: applying rss on queues 0-%d", + netdev_get_name(&dev->up), rss_n_rxq - 1); + } + + memset(&info, 0, sizeof(info)); + err = rte_eth_dev_info_get(dev->port_id, &info); + if (err < 0) { + goto out; + } + + if (info.reta_size % rss_n_rxq != 0 && + info.reta_size < RTE_ETH_RSS_RETA_SIZE_128) { + /* + * Some drivers set reta_size equal to the total number of rxqs that + * are configured when it is a power of two. Since we are actually + * reconfiguring the redirection table to exclude the last rxq, we may + * end up with an imbalanced redirection table. For example, such + * configuration: + * + * options:n_rxq=3 options:cp-protection=lacp + * + * Will actually configure 4 rxqs on the NIC, and the default reta to: + * + * [0, 1, 2, 3] + * + * And dpdk_cp_prot_rss_configure() will reconfigure reta to: + * + * [0, 1, 2, 0] + * + * Causing queue 0 to receive twice as much traffic as queues 1 and 2. + * + * Work around that corner case by forcing a bigger redirection table + * size to 128 entries when reta_size is not a multiple of rss_n_rxq + * and when reta_size is less than 128. This value seems to be + * supported by most of the drivers that also support rte flow. + */ + info.reta_size = RTE_ETH_RSS_RETA_SIZE_128; + } + + memset(reta_conf, 0, sizeof(reta_conf)); + for (uint16_t i = 0; i < info.reta_size; i++) { + uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE; + uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE; + reta_conf[idx].mask |= 1ULL << shift; + reta_conf[idx].reta[shift] = i % rss_n_rxq; + } + err = rte_eth_dev_rss_reta_update(dev->port_id, reta_conf, info.reta_size); + +out: + if (err < 0 && verbose) { + VLOG_DBG("%s: failed to configure RSS redirection table: err=%d", + netdev_get_name(&dev->up), err); + } + return err; +} + +static int +dpdk_cp_prot_configure(struct netdev_dpdk *dev, bool dry_run) +{ + int err = 0; + + if (dev->requested_cp_prot_flags & DPDK_CP_PROT_UNSUPPORTED) { + goto out; + } + if (dev->up.n_rxq < 2) { + err = ENOTSUP; + VLOG_DBG("%s: cp-protection: not enough available rx queues", + netdev_get_name(&dev->up)); + goto out; + } + + if (dev->requested_cp_prot_flags & DPDK_CP_PROT_LACP) { + err = dpdk_cp_prot_add_traffic_flow( + dev, + (const struct rte_flow_item []) { + { + .type = RTE_FLOW_ITEM_TYPE_ETH, + .spec = &(const struct rte_flow_item_eth){ + .type = htons(ETH_TYPE_LACP), + }, + .mask = &(const struct rte_flow_item_eth){ + .type = htons(0xffff), + }, + }, + { .type = RTE_FLOW_ITEM_TYPE_END }, + }, + "lacp", + dry_run + ); + if (err) { + goto out; + } + } + + if (!dry_run && dev->cp_prot_flows_num) { + /* reconfigure RSS reta in all but the cp protection queue */ + err = dpdk_cp_prot_rss_configure(dev, dev->up.n_rxq - 1, true); + } + +out: + if (!dry_run) { + dev->cp_prot_flags = dev->requested_cp_prot_flags; + } + if (err) { + dev->requested_cp_prot_flags |= DPDK_CP_PROT_UNSUPPORTED; + } + return err; +} + +static void +dpdk_cp_prot_unconfigure(struct netdev_dpdk *dev) +{ + struct rte_flow_error error; + + if (dev->cp_prot_flows_num == 0) { + return; + } + + VLOG_DBG("%s: cp-protection: reset flows", netdev_get_name(&dev->up)); + + for (int i = 0; i < dev->cp_prot_flows_num; i++) { + if (rte_flow_destroy(dev->port_id, dev->cp_prot_flows[i], &error)) { + VLOG_DBG("%s: cp-protection: failed to destroy flow: %s", + netdev_get_name(&dev->up), error.message); + } + } + free(dev->cp_prot_flows); + dev->cp_prot_flows_num = 0; + dev->cp_prot_flows = NULL; + + (void) dpdk_cp_prot_rss_configure(dev, dev->up.n_rxq, false); +} + static int netdev_dpdk_reconfigure(struct netdev *netdev) { @@ -4941,6 +5239,7 @@ netdev_dpdk_reconfigure(struct netdev *netdev) if (netdev->n_txq == dev->requested_n_txq && netdev->n_rxq == dev->requested_n_rxq + && dev->cp_prot_flags == dev->requested_cp_prot_flags && dev->mtu == dev->requested_mtu && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode && dev->rxq_size == dev->requested_rxq_size @@ -4968,6 +5267,8 @@ netdev_dpdk_reconfigure(struct netdev *netdev) goto out; } + dpdk_cp_prot_unconfigure(dev); + dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; netdev->n_txq = dev->requested_n_txq; @@ -5012,6 +5313,20 @@ netdev_dpdk_reconfigure(struct netdev *netdev) if (!dev->tx_q) { err = ENOMEM; } + if (!err && dev->requested_cp_prot_flags) { + /* dry run first */ + err = dpdk_cp_prot_configure(dev, true); + if (!err) { + /* if no error, apply configuration */ + err = dpdk_cp_prot_configure(dev, false); + } + if (err) { + /* no hw support, remove the extra queue & recover gracefully */ + err = 0; + dev->requested_n_rxq -= 1; + netdev_request_reconfigure(netdev); + } + } netdev_change_seq_changed(netdev); @@ -5213,7 +5528,13 @@ netdev_dpdk_flow_api_supported(struct netdev *netdev) ovs_mutex_lock(&dev->mutex); if (dev->type == DPDK_DEV_ETH) { /* TODO: Check if we able to offload some minimal flow. */ - ret = true; + if (dev->requested_cp_prot_flags || dev->cp_prot_flags) { + VLOG_WARN( + "%s: hw-offload is mutually exclusive with cp-protection", + netdev_get_name(netdev)); + } else { + ret = true; + } } ovs_mutex_unlock(&dev->mutex); out: diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml index f9bdb2d92bec..e34a1728108f 100644 --- a/vswitchd/vswitch.xml +++ b/vswitchd/vswitch.xml @@ -3430,6 +3430,32 @@ ovs-vsctl add-port br0 p0 -- set Interface p0 type=patch options:peer=p1 \ <p>This option may only be used with dpdk VF representors.</p> </column> + <column name="options" key="cp-protection" + type='{"type": "string", "enum": ["set", ["lacp"]]}'> + <p> + Allocate an extra Rx queue for control plane packets of the specified + protocol(s). + </p> + <p> + If the user has already configured multiple + <code>options:n_rxq</code> on the port, an additional one will be + allocated for control plane packets. If the hardware cannot satisfy + the requested number of requested Rx queues, the last Rx queue will + be assigned for control plane. If only one Rx queue is available or + if the hardware does not support the RTE flow matchers/actions + required to redirect the selected protocols, + <code>cp-protection</code> will be disabled. + </p> + <p> + This feature is multually exclusive with + <code>other_options:hw-offload</code> as it may conflict with the + offloaded RTE flows. + </p> + <p> + Disabled by default. + </p> + </column> + <column name="other_config" key="tx-steering" type='{"type": "string", "enum": ["set", ["thread", "hash"]]}'>
Some control protocols are used to maintain link status between forwarding engines (e.g. LACP). When the system is not sized properly, the PMD threads may not be able to process all incoming traffic from the configured Rx queues. When a signaling packet of such protocols is dropped, it can cause link flapping, worsening the situation. Use the RTE flow API to redirect these protocols into a dedicated Rx queue. The assumption is made that the ratio between control protocol traffic and user data traffic is very low and thus this dedicated Rx queue will never get full. The RSS redirection table is re-programmed to only use the other Rx queues. The RSS table size is stored in the netdev_dpdk structure at port initialization to avoid requesting the information again when changing the port configuration. The additional Rx queue will be assigned a PMD core like any other Rx queue. Polling that extra queue may introduce increased latency and a slight performance penalty at the benefit of preventing link flapping. This feature must be enabled per port on specific protocols via the cp-protection option. This option takes a coma-separated list of protocol names. It is only supported on ethernet ports. If the user has already configured multiple Rx queues on the port, an additional one will be allocated for control plane packets. If the hardware cannot satisfy the requested number of requested Rx queues, the last Rx queue will be assigned for control plane. If only one Rx queue is available, the cp-protection feature will be disabled. If the hardware does not support the RTE flow matchers/actions, the feature will be disabled. It cannot be enabled when other_config:hw-offload=true as it may conflict with the offloaded RTE flows. Similarly, if hw-offload is enabled while some ports already have cp-protection enabled, the RTE flow offloading will be disabled on these ports. Example use: ovs-vsctl add-bond br-phy bond0 phy0 phy1 -- \ set interface phy0 type=dpdk options:dpdk-devargs=0000:ca:00.0 -- \ set interface phy0 options:cp-protection=lacp -- \ set interface phy1 type=dpdk options:dpdk-devargs=0000:ca:00.1 -- \ set interface phy1 options:cp-protection=lacp As a starting point, only one protocol is supported: LACP. Other protocols can be added in the future. NIC compatibility should be checked. To validate that this works as intended, I used a traffic generator to generate random traffic slightly above the machine capacity at line rate on a two ports bond interface. OVS is configured to receive traffic on two VLANs and pop/push them in a br-int bridge based on tags set on patch ports. +----------------------+ | DUT | |+--------------------+| || br-int || default flow, action=NORMAL || || || patch10 patch11 || |+---|-----------|----+| | | | | |+---|-----------|----+| || patch00 patch01 || || tag:10 tag:20 || || || || br-phy || default flow, action=NORMAL || || || bond0 || balance-slb, lacp=passive, lacp-time=fast || phy0 phy1 || |+------|-----|-------+| +-------|-----|--------+ | | +-------|-----|--------+ | port0 port1 | balance L3/L4, lacp=active, lacp-time=fast | lag | mode trunk VLANs 10, 20 | | | switch | | | | vlan 10 vlan 20 | mode access | port2 port3 | +-----|----------|-----+ | | +-----|----------|-----+ | port0 port1 | Random traffic that is properly balanced | | across the bond ports in both directions. | traffic generator | +----------------------+ Without cp-protection, the bond0 links are randomly switching to "defaulted" when one of the LACP packets sent by the switch is dropped because the RX queues are full and the PMD threads did not process them fast enough. When that happens, all traffic must go through a single link which causes above line rate traffic to be dropped. When cp-protection is enabled, no LACP packet is dropped and the bond links remain enabled at all times, maximizing the throughput. This feature may be considered as "QoS". However, it does not work by limiting the rate of traffic explicitly. It only guarantees that some protocols have a lower chance of being dropped because the PMD cores cannot keep up with regular traffic. The choice of protocols is limited on purpose. This is not meant to be configurable by users. Some limited configurability could be considered in the future but it would expose to more potential issues if users are accidentally redirecting all traffic in the control plane queue. Cc: Christophe Fontaine <cfontain@redhat.com> Cc: Kevin Traynor <ktraynor@redhat.com> Signed-off-by: Robin Jarry <rjarry@redhat.com> --- v3 -> v4: * Rebased on master 739bcf2263b3 ("odp-execute: Fix ipv4 missing clearing of connection tracking fields.") * Removed the extra reta_size field from struct netdev_dpdk. rte_eth_dev_info() is now called every time in dpdk_cp_prot_rss_configure(). * Fixed RSS reconfiguration for drivers that have redirection tables that are smaller than RTE_ETH_RETA_GROUP_SIZE (64). * Added workaround to avoid imbalanced queues when the reta size is too small and is not a multiple of the number of RSS queues. * Fixed panic when deleting ports with cp-protection enabled. * Using cp-protection with an invalid/unsupported protocol now only prints a warning in the logs and it is ignored. It no longer causes the port to be destroyed. Documentation/topics/dpdk/phy.rst | 55 +++++ lib/netdev-dpdk.c | 323 +++++++++++++++++++++++++++++- vswitchd/vswitch.xml | 26 +++ 3 files changed, 403 insertions(+), 1 deletion(-)