net/sfc: implement representor Rx
authorIgor Romanov <igor.romanov@oktetlabs.ru>
Mon, 11 Oct 2021 14:48:44 +0000 (17:48 +0300)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 12 Oct 2021 16:44:11 +0000 (18:44 +0200)
Implement traffic forwarding for representor and representor proxy
from virtual functions to representor Rx queues.

Signed-off-by: Igor Romanov <igor.romanov@oktetlabs.ru>
Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Reviewed-by: Andy Moreton <amoreton@xilinx.com>
Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
drivers/net/sfc/sfc_repr.c
drivers/net/sfc/sfc_repr_proxy.c
drivers/net/sfc/sfc_repr_proxy.h

index fc3229a..64753f2 100644 (file)
@@ -160,6 +160,16 @@ sfc_repr_tx_queue_stop(void *queue)
        rte_ring_reset(txq->ring);
 }
 
+static uint16_t
+sfc_repr_rx_burst(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+       struct sfc_repr_rxq *rxq = rx_queue;
+       void **objs = (void *)&rx_pkts[0];
+
+       /* mbufs port is already filled correctly by representors proxy */
+       return rte_ring_sc_dequeue_burst(rxq->ring, objs, nb_pkts, NULL);
+}
+
 static uint16_t
 sfc_repr_tx_burst(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
@@ -803,6 +813,7 @@ sfc_repr_dev_close(struct rte_eth_dev *dev)
 
        (void)sfc_repr_proxy_del_port(srs->pf_port_id, srs->repr_id);
 
+       dev->rx_pkt_burst = NULL;
        dev->tx_pkt_burst = NULL;
        dev->dev_ops = NULL;
 
@@ -924,6 +935,7 @@ sfc_repr_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
                goto fail_mac_addrs;
        }
 
+       dev->rx_pkt_burst = sfc_repr_rx_burst;
        dev->tx_pkt_burst = sfc_repr_tx_burst;
        dev->dev_ops = &sfc_repr_dev_ops;
 
index d8934ba..535b07e 100644 (file)
@@ -18,6 +18,7 @@
 #include "sfc_ev.h"
 #include "sfc_rx.h"
 #include "sfc_tx.h"
+#include "sfc_dp_rx.h"
 
 /**
  * Amount of time to wait for the representor proxy routine (which is
@@ -31,6 +32,8 @@
  */
 #define SFC_REPR_PROXY_ROUTINE_TERMINATE_TIMEOUT_MS    10000
 
+#define SFC_REPR_INVALID_ROUTE_PORT_ID  (UINT16_MAX)
+
 static struct sfc_repr_proxy *
 sfc_repr_proxy_by_adapter(struct sfc_adapter *sa)
 {
@@ -187,6 +190,113 @@ sfc_repr_proxy_handle_tx(struct sfc_repr_proxy_dp_txq *rp_txq,
        }
 }
 
+static struct sfc_repr_proxy_port *
+sfc_repr_proxy_rx_route_mbuf(struct sfc_repr_proxy *rp, struct rte_mbuf *m)
+{
+       struct sfc_repr_proxy_port *port;
+       efx_mport_id_t mport_id;
+
+       mport_id.id = *RTE_MBUF_DYNFIELD(m, sfc_dp_mport_offset,
+                                        typeof(&((efx_mport_id_t *)0)->id));
+
+       TAILQ_FOREACH(port, &rp->ports, entries) {
+               if (port->egress_mport.id == mport_id.id) {
+                       m->port = port->rte_port_id;
+                       m->ol_flags &= ~sfc_dp_mport_override;
+                       return port;
+               }
+       }
+
+       return NULL;
+}
+
+/*
+ * Returns true if a packet is encountered which should be forwarded to a
+ * port which is different from the one that is currently routed.
+ */
+static bool
+sfc_repr_proxy_rx_route(struct sfc_repr_proxy *rp,
+                       struct sfc_repr_proxy_dp_rxq *rp_rxq)
+{
+       unsigned int i;
+
+       for (i = rp_rxq->routed;
+            i < rp_rxq->available && !rp_rxq->stop_route;
+            i++, rp_rxq->routed++) {
+               struct sfc_repr_proxy_port *port;
+               struct rte_mbuf *m = rp_rxq->pkts[i];
+
+               port = sfc_repr_proxy_rx_route_mbuf(rp, m);
+               /* Cannot find destination representor */
+               if (port == NULL) {
+                       /* Effectively drop the packet */
+                       rp_rxq->forwarded++;
+                       continue;
+               }
+
+               /* Currently routed packets are mapped to a different port */
+               if (port->repr_id != rp_rxq->route_port_id &&
+                   rp_rxq->route_port_id != SFC_REPR_INVALID_ROUTE_PORT_ID)
+                       return true;
+
+               rp_rxq->route_port_id = port->repr_id;
+       }
+
+       return false;
+}
+
+static void
+sfc_repr_proxy_rx_forward(struct sfc_repr_proxy *rp,
+                         struct sfc_repr_proxy_dp_rxq *rp_rxq)
+{
+       struct sfc_repr_proxy_port *port;
+
+       if (rp_rxq->route_port_id != SFC_REPR_INVALID_ROUTE_PORT_ID) {
+               port = sfc_repr_proxy_find_port(rp, rp_rxq->route_port_id);
+
+               if (port != NULL && port->started) {
+                       rp_rxq->forwarded +=
+                           rte_ring_sp_enqueue_burst(port->rxq[0].ring,
+                               (void **)(&rp_rxq->pkts[rp_rxq->forwarded]),
+                               rp_rxq->routed - rp_rxq->forwarded, NULL);
+               } else {
+                       /* Drop all routed packets if the port is not started */
+                       rp_rxq->forwarded = rp_rxq->routed;
+               }
+       }
+
+       if (rp_rxq->forwarded == rp_rxq->routed) {
+               rp_rxq->route_port_id = SFC_REPR_INVALID_ROUTE_PORT_ID;
+               rp_rxq->stop_route = false;
+       } else {
+               /* Stall packet routing if not all packets were forwarded */
+               rp_rxq->stop_route = true;
+       }
+
+       if (rp_rxq->available == rp_rxq->forwarded)
+               rp_rxq->available = rp_rxq->forwarded = rp_rxq->routed = 0;
+}
+
+static void
+sfc_repr_proxy_handle_rx(struct sfc_repr_proxy *rp,
+                        struct sfc_repr_proxy_dp_rxq *rp_rxq)
+{
+       bool route_again;
+
+       if (rp_rxq->available < RTE_DIM(rp_rxq->pkts)) {
+               rp_rxq->available += rp_rxq->pkt_burst(rp_rxq->dp,
+                               &rp_rxq->pkts[rp_rxq->available],
+                               RTE_DIM(rp_rxq->pkts) - rp_rxq->available);
+               if (rp_rxq->available == rp_rxq->forwarded)
+                       return;
+       }
+
+       do {
+               route_again = sfc_repr_proxy_rx_route(rp, rp_rxq);
+               sfc_repr_proxy_rx_forward(rp, rp_rxq);
+       } while (route_again && !rp_rxq->stop_route);
+}
+
 static int32_t
 sfc_repr_proxy_routine(void *arg)
 {
@@ -204,6 +314,9 @@ sfc_repr_proxy_routine(void *arg)
                        sfc_repr_proxy_handle_tx(&rp->dp_txq[i], &port->txq[i]);
        }
 
+       for (i = 0; i < rp->nb_rxq; i++)
+               sfc_repr_proxy_handle_rx(rp, &rp->dp_rxq[i]);
+
        return 0;
 }
 
@@ -412,6 +525,18 @@ sfc_repr_proxy_rxq_detach(struct sfc_adapter *sa)
        sfc_log_init(sa, "done");
 }
 
+static struct sfc_rxq_info *
+sfc_repr_proxy_rxq_info_get(struct sfc_adapter *sa, unsigned int repr_queue_id)
+{
+       struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
+       struct sfc_repr_proxy_dp_rxq *dp_rxq;
+
+       SFC_ASSERT(repr_queue_id < sfc_repr_nb_rxq(sas));
+       dp_rxq = &sa->repr_proxy.dp_rxq[repr_queue_id];
+
+       return &sas->rxq_info[dp_rxq->sw_index];
+}
+
 static int
 sfc_repr_proxy_rxq_init(struct sfc_adapter *sa,
                        struct sfc_repr_proxy_dp_rxq *rxq)
@@ -539,6 +664,14 @@ sfc_repr_proxy_rxq_start(struct sfc_adapter *sa)
                                i);
                        goto fail_start;
                }
+
+               rxq->dp = sfc_repr_proxy_rxq_info_get(sa, i)->dp;
+               rxq->pkt_burst = sa->eth_dev->rx_pkt_burst;
+               rxq->available = 0;
+               rxq->routed = 0;
+               rxq->forwarded = 0;
+               rxq->stop_route = false;
+               rxq->route_port_id = SFC_REPR_INVALID_ROUTE_PORT_ID;
        }
 
        sfc_log_init(sa, "done");
@@ -993,6 +1126,7 @@ sfc_repr_proxy_start(struct sfc_adapter *sa)
                goto fail_txq_start;
 
        rp->nb_txq = sfc_repr_nb_txq(sas);
+       rp->nb_rxq = sfc_repr_nb_rxq(sas);
 
        /* Service core may be in "stopped" state, start it */
        rc = rte_service_lcore_start(rp->service_core_id);
index d47e0a4..b49b1a2 100644 (file)
@@ -75,6 +75,16 @@ struct sfc_repr_proxy_dp_rxq {
        struct rte_mempool              *mp;
        unsigned int                    ref_count;
 
+       eth_rx_burst_t                  pkt_burst;
+       struct sfc_dp_rxq               *dp;
+
+       uint16_t                        route_port_id;
+       bool                            stop_route;
+       unsigned int                    available;
+       unsigned int                    forwarded;
+       unsigned int                    routed;
+       struct rte_mbuf                 *pkts[SFC_REPR_PROXY_TX_BURST];
+
        sfc_sw_index_t                  sw_index;
 };
 
@@ -118,6 +128,7 @@ struct sfc_repr_proxy {
 
        struct sfc_repr_proxy_mbox      mbox;
        unsigned int                    nb_txq;
+       unsigned int                    nb_rxq;
 };
 
 struct sfc_adapter;