Merge branch 'sfc-vf-representors-for-ef100-rx-side'

Edward Cree says:

====================
sfc: VF representors for EF100 - RX side

This series adds the receive path for EF100 VF representors, plus other
 minor features such as statistics.
====================

Link: https://lore.kernel.org/r/cover.1659034549.git.ecree.xilinx@gmail.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/sfc/Makefile

@@ -8,7 +8,8 @@ sfc-y			+= efx.o efx_common.o efx_channels.o nic.o \
 			   ef100.o ef100_nic.o ef100_netdev.o \
 			   ef100_ethtool.o ef100_rx.o ef100_tx.o
 sfc-$(CONFIG_SFC_MTD)	+= mtd.o
-sfc-$(CONFIG_SFC_SRIOV)	+= sriov.o ef10_sriov.o ef100_sriov.o ef100_rep.o mae.o
+sfc-$(CONFIG_SFC_SRIOV)	+= sriov.o ef10_sriov.o ef100_sriov.o ef100_rep.o \
+                           mae.o tc.o
 
 obj-$(CONFIG_SFC)	+= sfc.o
 

drivers/net/ethernet/sfc/ef10.c

@@ -2538,23 +2538,33 @@ static int efx_ef10_filter_table_probe(struct efx_nic *efx)
 
 	if (rc)
 		return rc;
+	down_write(&efx->filter_sem);
 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
 
 	if (rc)
-		return rc;
+		goto out_unlock;
 
 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
 		if (rc)
 			goto fail_add_vlan;
 	}
-	return 0;
+	goto out_unlock;
 
 fail_add_vlan:
 	efx_mcdi_filter_table_remove(efx);
+out_unlock:
+	up_write(&efx->filter_sem);
 	return rc;
 }
 
+static void efx_ef10_filter_table_remove(struct efx_nic *efx)
+{
+	down_write(&efx->filter_sem);
+	efx_mcdi_filter_table_remove(efx);
+	up_write(&efx->filter_sem);
+}
+
 /* This creates an entry in the RX descriptor queue */
 static inline void
 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
@@ -3211,9 +3221,7 @@ static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
 
 	efx_device_detach_sync(efx);
 	efx_net_stop(efx->net_dev);
-	down_write(&efx->filter_sem);
-	efx_mcdi_filter_table_remove(efx);
-	up_write(&efx->filter_sem);
+	efx_ef10_filter_table_remove(efx);
 
 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
 	if (rc)
@@ -3243,9 +3251,7 @@ static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
 	if (rc2)
 		goto reset_nic;
 restore_filters:
-	down_write(&efx->filter_sem);
 	rc2 = efx_ef10_filter_table_probe(efx);
-	up_write(&efx->filter_sem);
 	if (rc2)
 		goto reset_nic;
 
@@ -3275,8 +3281,7 @@ static int efx_ef10_set_mac_address(struct efx_nic *efx)
 	efx_net_stop(efx->net_dev);
 
 	mutex_lock(&efx->mac_lock);
-	down_write(&efx->filter_sem);
-	efx_mcdi_filter_table_remove(efx);
+	efx_ef10_filter_table_remove(efx);
 
 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
 			efx->net_dev->dev_addr);
@@ -3286,7 +3291,6 @@ static int efx_ef10_set_mac_address(struct efx_nic *efx)
 				sizeof(inbuf), NULL, 0, NULL);
 
 	efx_ef10_filter_table_probe(efx);
-	up_write(&efx->filter_sem);
 	mutex_unlock(&efx->mac_lock);
 
 	if (was_enabled)
@@ -4092,7 +4096,7 @@ const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
 	.ev_test_generate = efx_ef10_ev_test_generate,
 	.filter_table_probe = efx_ef10_filter_table_probe,
 	.filter_table_restore = efx_mcdi_filter_table_restore,
-	.filter_table_remove = efx_mcdi_filter_table_remove,
+	.filter_table_remove = efx_ef10_filter_table_remove,
 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
 	.filter_insert = efx_mcdi_filter_insert,
 	.filter_remove_safe = efx_mcdi_filter_remove_safe,

drivers/net/ethernet/sfc/ef100.c

@@ -431,6 +431,9 @@ static void ef100_pci_remove(struct pci_dev *pci_dev)
 
 	probe_data = container_of(efx, struct efx_probe_data, efx);
 	ef100_remove_netdev(probe_data);
+#ifdef CONFIG_SFC_SRIOV
+	efx_fini_struct_tc(efx);
+#endif
 
 	ef100_remove(efx);
 	efx_fini_io(efx);

drivers/net/ethernet/sfc/ef100_netdev.c

@@ -329,6 +329,10 @@ void ef100_remove_netdev(struct efx_probe_data *probe_data)
 
 	ef100_unregister_netdev(efx);
 
+#ifdef CONFIG_SFC_SRIOV
+	efx_fini_tc(efx);
+#endif
+
 	down_write(&efx->filter_sem);
 	efx_mcdi_filter_table_remove(efx);
 	up_write(&efx->filter_sem);

drivers/net/ethernet/sfc/ef100_nic.c

@@ -24,6 +24,8 @@
 #include "ef100_tx.h"
 #include "ef100_sriov.h"
 #include "ef100_netdev.h"
+#include "tc.h"
+#include "mae.h"
 #include "rx_common.h"
 
 #define EF100_MAX_VIS 4096
@@ -374,26 +376,46 @@ static int ef100_filter_table_up(struct efx_nic *efx)
 {
 	int rc;
 
+	down_write(&efx->filter_sem);
 	rc = efx_mcdi_filter_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
-	if (rc) {
-		efx_mcdi_filter_table_down(efx);
-		return rc;
-	}
+	if (rc)
+		goto fail_unspec;
 
 	rc = efx_mcdi_filter_add_vlan(efx, 0);
-	if (rc) {
-		efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC);
-		efx_mcdi_filter_table_down(efx);
-	}
+	if (rc)
+		goto fail_vlan0;
+	/* Drop the lock: we've finished altering table existence, and
+	 * filter insertion will need to take the lock for read.
+	 */
+	up_write(&efx->filter_sem);
+#ifdef CONFIG_SFC_SRIOV
+	rc = efx_tc_insert_rep_filters(efx);
+	/* Rep filter failure is nonfatal */
+	if (rc)
+		netif_warn(efx, drv, efx->net_dev,
+			   "Failed to insert representor filters, rc %d\n",
+			   rc);
+#endif
+	return 0;
 
+fail_vlan0:
+	efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC);
+fail_unspec:
+	efx_mcdi_filter_table_down(efx);
+	up_write(&efx->filter_sem);
 	return rc;
 }
 
 static void ef100_filter_table_down(struct efx_nic *efx)
 {
+#ifdef CONFIG_SFC_SRIOV
+	efx_tc_remove_rep_filters(efx);
+#endif
+	down_write(&efx->filter_sem);
 	efx_mcdi_filter_del_vlan(efx, 0);
 	efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC);
 	efx_mcdi_filter_table_down(efx);
+	up_write(&efx->filter_sem);
 }
 
 /*	Other
@@ -704,6 +726,31 @@ static unsigned int efx_ef100_recycle_ring_size(const struct efx_nic *efx)
 	return 10 * EFX_RECYCLE_RING_SIZE_10G;
 }
 
+#ifdef CONFIG_SFC_SRIOV
+static int efx_ef100_get_base_mport(struct efx_nic *efx)
+{
+	struct ef100_nic_data *nic_data = efx->nic_data;
+	u32 selector, id;
+	int rc;
+
+	/* Construct mport selector for "physical network port" */
+	efx_mae_mport_wire(efx, &selector);
+	/* Look up actual mport ID */
+	rc = efx_mae_lookup_mport(efx, selector, &id);
+	if (rc)
+		return rc;
+	/* The ID should always fit in 16 bits, because that's how wide the
+	 * corresponding fields in the RX prefix & TX override descriptor are
+	 */
+	if (id >> 16)
+		netif_warn(efx, probe, efx->net_dev, "Bad base m-port id %#x\n",
+			   id);
+	nic_data->base_mport = id;
+	nic_data->have_mport = true;
+	return 0;
+}
+#endif
+
 static int compare_versions(const char *a, const char *b)
 {
 	int a_major, a_minor, a_point, a_patch;
@@ -1064,6 +1111,34 @@ int ef100_probe_netdev_pf(struct efx_nic *efx)
 	eth_hw_addr_set(net_dev, net_dev->perm_addr);
 	memcpy(nic_data->port_id, net_dev->perm_addr, ETH_ALEN);
 
+	if (!nic_data->grp_mae)
+		return 0;
+
+#ifdef CONFIG_SFC_SRIOV
+	rc = efx_init_struct_tc(efx);
+	if (rc)
+		return rc;
+
+	rc = efx_ef100_get_base_mport(efx);
+	if (rc) {
+		netif_warn(efx, probe, net_dev,
+			   "Failed to probe base mport rc %d; representors will not function\n",
+			   rc);
+	}
+
+	rc = efx_init_tc(efx);
+	if (rc) {
+		/* Either we don't have an MAE at all (i.e. legacy v-switching),
+		 * or we do but we failed to probe it.  In the latter case, we
+		 * may not have set up default rules, in which case we won't be
+		 * able to pass any traffic.  However, we don't fail the probe,
+		 * because the user might need to use the netdevice to apply
+		 * configuration changes to fix whatever's wrong with the MAE.
+		 */
+		netif_warn(efx, probe, net_dev, "Failed to probe MAE rc %d\n",
+			   rc);
+	}
+#endif
 	return 0;
 
 fail:

drivers/net/ethernet/sfc/ef100_nic.h

@@ -72,6 +72,8 @@ struct ef100_nic_data {
 	u8 port_id[ETH_ALEN];
 	DECLARE_BITMAP(evq_phases, EFX_MAX_CHANNELS);
 	u64 stats[EF100_STAT_COUNT];
+	u32 base_mport;
+	bool have_mport; /* base_mport was populated successfully */
 	bool grp_mae; /* MAE Privilege */
 	u16 tso_max_hdr_len;
 	u16 tso_max_payload_num_segs;

drivers/net/ethernet/sfc/ef100_rep.c

@@ -13,15 +13,24 @@
 #include "ef100_netdev.h"
 #include "ef100_nic.h"
 #include "mae.h"
+#include "rx_common.h"
 
 #define EFX_EF100_REP_DRIVER	"efx_ef100_rep"
 
+#define EFX_REP_DEFAULT_PSEUDO_RING_SIZE	64
+
+static int efx_ef100_rep_poll(struct napi_struct *napi, int weight);
+
 static int efx_ef100_rep_init_struct(struct efx_nic *efx, struct efx_rep *efv,
 				     unsigned int i)
 {
 	efv->parent = efx;
 	efv->idx = i;
 	INIT_LIST_HEAD(&efv->list);
+	efv->dflt.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+	INIT_LIST_HEAD(&efv->dflt.acts.list);
+	INIT_LIST_HEAD(&efv->rx_list);
+	spin_lock_init(&efv->rx_lock);
 	efv->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE |
 			  NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
 			  NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
@@ -29,6 +38,25 @@ static int efx_ef100_rep_init_struct(struct efx_nic *efx, struct efx_rep *efv,
 	return 0;
 }
 
+static int efx_ef100_rep_open(struct net_device *net_dev)
+{
+	struct efx_rep *efv = netdev_priv(net_dev);
+
+	netif_napi_add(net_dev, &efv->napi, efx_ef100_rep_poll,
+		       NAPI_POLL_WEIGHT);
+	napi_enable(&efv->napi);
+	return 0;
+}
+
+static int efx_ef100_rep_close(struct net_device *net_dev)
+{
+	struct efx_rep *efv = netdev_priv(net_dev);
+
+	napi_disable(&efv->napi);
+	netif_napi_del(&efv->napi);
+	return 0;
+}
+
 static netdev_tx_t efx_ef100_rep_xmit(struct sk_buff *skb,
 				      struct net_device *dev)
 {
@@ -79,10 +107,26 @@ static int efx_ef100_rep_get_phys_port_name(struct net_device *dev,
 	return 0;
 }
 
+static void efx_ef100_rep_get_stats64(struct net_device *dev,
+				      struct rtnl_link_stats64 *stats)
+{
+	struct efx_rep *efv = netdev_priv(dev);
+
+	stats->rx_packets = atomic64_read(&efv->stats.rx_packets);
+	stats->tx_packets = atomic64_read(&efv->stats.tx_packets);
+	stats->rx_bytes = atomic64_read(&efv->stats.rx_bytes);
+	stats->tx_bytes = atomic64_read(&efv->stats.tx_bytes);
+	stats->rx_dropped = atomic64_read(&efv->stats.rx_dropped);
+	stats->tx_errors = atomic64_read(&efv->stats.tx_errors);
+}
+
 static const struct net_device_ops efx_ef100_rep_netdev_ops = {
+	.ndo_open		= efx_ef100_rep_open,
+	.ndo_stop		= efx_ef100_rep_close,
 	.ndo_start_xmit		= efx_ef100_rep_xmit,
 	.ndo_get_port_parent_id	= efx_ef100_rep_get_port_parent_id,
 	.ndo_get_phys_port_name	= efx_ef100_rep_get_phys_port_name,
+	.ndo_get_stats64	= efx_ef100_rep_get_stats64,
 };
 
 static void efx_ef100_rep_get_drvinfo(struct net_device *dev,
@@ -106,10 +150,37 @@ static void efx_ef100_rep_ethtool_set_msglevel(struct net_device *net_dev,
 	efv->msg_enable = msg_enable;
 }
 
+static void efx_ef100_rep_ethtool_get_ringparam(struct net_device *net_dev,
+						struct ethtool_ringparam *ring,
+						struct kernel_ethtool_ringparam *kring,
+						struct netlink_ext_ack *ext_ack)
+{
+	struct efx_rep *efv = netdev_priv(net_dev);
+
+	ring->rx_max_pending = U32_MAX;
+	ring->rx_pending = efv->rx_pring_size;
+}
+
+static int efx_ef100_rep_ethtool_set_ringparam(struct net_device *net_dev,
+					       struct ethtool_ringparam *ring,
+					       struct kernel_ethtool_ringparam *kring,
+					       struct netlink_ext_ack *ext_ack)
+{
+	struct efx_rep *efv = netdev_priv(net_dev);
+
+	if (ring->rx_mini_pending || ring->rx_jumbo_pending || ring->tx_pending)
+		return -EINVAL;
+
+	efv->rx_pring_size = ring->rx_pending;
+	return 0;
+}
+
 static const struct ethtool_ops efx_ef100_rep_ethtool_ops = {
 	.get_drvinfo		= efx_ef100_rep_get_drvinfo,
 	.get_msglevel		= efx_ef100_rep_ethtool_get_msglevel,
 	.set_msglevel		= efx_ef100_rep_ethtool_set_msglevel,
+	.get_ringparam		= efx_ef100_rep_ethtool_get_ringparam,
+	.set_ringparam		= efx_ef100_rep_ethtool_set_ringparam,
 };
 
 static struct efx_rep *efx_ef100_rep_create_netdev(struct efx_nic *efx,
@@ -159,6 +230,7 @@ static int efx_ef100_configure_rep(struct efx_rep *efv)
 	u32 selector;
 	int rc;
 
+	efv->rx_pring_size = EFX_REP_DEFAULT_PSEUDO_RING_SIZE;
 	/* Construct mport selector for corresponding VF */
 	efx_mae_mport_vf(efx, efv->idx, &selector);
 	/* Look up actual mport ID */
@@ -169,7 +241,14 @@ static int efx_ef100_configure_rep(struct efx_rep *efv)
 	/* mport label should fit in 16 bits */
 	WARN_ON(efv->mport >> 16);
 
-	return 0;
+	return efx_tc_configure_default_rule_rep(efv);
+}
+
+static void efx_ef100_deconfigure_rep(struct efx_rep *efv)
+{
+	struct efx_nic *efx = efv->parent;
+
+	efx_tc_deconfigure_default_rule(efx, &efv->dflt);
 }
 
 static void efx_ef100_rep_destroy_netdev(struct efx_rep *efv)
@@ -181,6 +260,7 @@ static void efx_ef100_rep_destroy_netdev(struct efx_rep *efv)
 	list_del(&efv->list);
 	spin_unlock_bh(&efx->vf_reps_lock);
 	rtnl_unlock();
+	synchronize_rcu();
 	free_netdev(efv->net_dev);
 }
 
@@ -202,19 +282,21 @@ int efx_ef100_vfrep_create(struct efx_nic *efx, unsigned int i)
 		pci_err(efx->pci_dev,
 			"Failed to configure representor for VF %d, rc %d\n",
 			i, rc);
-		goto fail;
+		goto fail1;
 	}
 	rc = register_netdev(efv->net_dev);
 	if (rc) {
 		pci_err(efx->pci_dev,
 			"Failed to register representor for VF %d, rc %d\n",
 			i, rc);
-		goto fail;
+		goto fail2;
 	}
 	pci_dbg(efx->pci_dev, "Representor for VF %d is %s\n", i,
 		efv->net_dev->name);
 	return 0;
-fail:
+fail2:
+	efx_ef100_deconfigure_rep(efv);
+fail1:
 	efx_ef100_rep_destroy_netdev(efv);
 	return rc;
 }
@@ -228,6 +310,7 @@ void efx_ef100_vfrep_destroy(struct efx_nic *efx, struct efx_rep *efv)
 		return;
 	netif_dbg(efx, drv, rep_dev, "Removing VF representor\n");
 	unregister_netdev(rep_dev);
+	efx_ef100_deconfigure_rep(efv);
 	efx_ef100_rep_destroy_netdev(efv);
 }
 
@@ -242,3 +325,111 @@ void efx_ef100_fini_vfreps(struct efx_nic *efx)
 	list_for_each_entry_safe(efv, next, &efx->vf_reps, list)
 		efx_ef100_vfrep_destroy(efx, efv);
 }
+
+static int efx_ef100_rep_poll(struct napi_struct *napi, int weight)
+{
+	struct efx_rep *efv = container_of(napi, struct efx_rep, napi);
+	unsigned int read_index;
+	struct list_head head;
+	struct sk_buff *skb;
+	bool need_resched;
+	int spent = 0;
+
+	INIT_LIST_HEAD(&head);
+	/* Grab up to 'weight' pending SKBs */
+	spin_lock_bh(&efv->rx_lock);
+	read_index = efv->write_index;
+	while (spent < weight && !list_empty(&efv->rx_list)) {
+		skb = list_first_entry(&efv->rx_list, struct sk_buff, list);
+		list_del(&skb->list);
+		list_add_tail(&skb->list, &head);
+		spent++;
+	}
+	spin_unlock_bh(&efv->rx_lock);
+	/* Receive them */
+	netif_receive_skb_list(&head);
+	if (spent < weight)
+		if (napi_complete_done(napi, spent)) {
+			spin_lock_bh(&efv->rx_lock);
+			efv->read_index = read_index;
+			/* If write_index advanced while we were doing the
+			 * RX, then storing our read_index won't re-prime the
+			 * fake-interrupt.  In that case, we need to schedule
+			 * NAPI again to consume the additional packet(s).
+			 */
+			need_resched = efv->write_index != read_index;
+			spin_unlock_bh(&efv->rx_lock);
+			if (need_resched)
+				napi_schedule(&efv->napi);
+		}
+	return spent;
+}
+
+void efx_ef100_rep_rx_packet(struct efx_rep *efv, struct efx_rx_buffer *rx_buf)
+{
+	u8 *eh = efx_rx_buf_va(rx_buf);
+	struct sk_buff *skb;
+	bool primed;
+
+	/* Don't allow too many queued SKBs to build up, as they consume
+	 * GFP_ATOMIC memory.  If we overrun, just start dropping.
+	 */
+	if (efv->write_index - READ_ONCE(efv->read_index) > efv->rx_pring_size) {
+		atomic64_inc(&efv->stats.rx_dropped);
+		if (net_ratelimit())
+			netif_dbg(efv->parent, rx_err, efv->net_dev,
+				  "nodesc-dropped packet of length %u\n",
+				  rx_buf->len);
+		return;
+	}
+
+	skb = netdev_alloc_skb(efv->net_dev, rx_buf->len);
+	if (!skb) {
+		atomic64_inc(&efv->stats.rx_dropped);
+		if (net_ratelimit())
+			netif_dbg(efv->parent, rx_err, efv->net_dev,
+				  "noskb-dropped packet of length %u\n",
+				  rx_buf->len);
+		return;
+	}
+	memcpy(skb->data, eh, rx_buf->len);
+	__skb_put(skb, rx_buf->len);
+
+	skb_record_rx_queue(skb, 0); /* rep is single-queue */
+
+	/* Move past the ethernet header */
+	skb->protocol = eth_type_trans(skb, efv->net_dev);
+
+	skb_checksum_none_assert(skb);
+
+	atomic64_inc(&efv->stats.rx_packets);
+	atomic64_add(rx_buf->len, &efv->stats.rx_bytes);
+
+	/* Add it to the rx list */
+	spin_lock_bh(&efv->rx_lock);
+	primed = efv->read_index == efv->write_index;
+	list_add_tail(&skb->list, &efv->rx_list);
+	efv->write_index++;
+	spin_unlock_bh(&efv->rx_lock);
+	/* Trigger rx work */
+	if (primed)
+		napi_schedule(&efv->napi);
+}
+
+struct efx_rep *efx_ef100_find_rep_by_mport(struct efx_nic *efx, u16 mport)
+{
+	struct efx_rep *efv, *out = NULL;
+
+	/* spinlock guards against list mutation while we're walking it;
+	 * but caller must also hold rcu_read_lock() to ensure the netdev
+	 * isn't freed after we drop the spinlock.
+	 */
+	spin_lock_bh(&efx->vf_reps_lock);
+	list_for_each_entry(efv, &efx->vf_reps, list)
+		if (efv->mport == mport) {
+			out = efv;
+			break;
+		}
+	spin_unlock_bh(&efx->vf_reps_lock);
+	return out;
+}

drivers/net/ethernet/sfc/ef100_rep.h

@@ -14,6 +14,7 @@
 #define EF100_REP_H
 
 #include "net_driver.h"
+#include "tc.h"
 
 struct efx_rep_sw_stats {
 	atomic64_t rx_packets, tx_packets;
@@ -29,7 +30,14 @@ struct efx_rep_sw_stats {
  * @msg_enable: log message enable flags
  * @mport: m-port ID of corresponding VF
  * @idx: VF index
+ * @write_index: number of packets enqueued to @rx_list
+ * @read_index: number of packets consumed from @rx_list
+ * @rx_pring_size: max length of RX list
+ * @dflt: default-rule for MAE switching
  * @list: entry on efx->vf_reps
+ * @rx_list: list of SKBs queued for receive in NAPI poll
+ * @rx_lock: protects @rx_list
+ * @napi: NAPI control structure
  * @stats: software traffic counters for netdev stats
  */
 struct efx_rep {
@@ -38,7 +46,13 @@ struct efx_rep {
 	u32 msg_enable;
 	u32 mport;
 	unsigned int idx;
+	unsigned int write_index, read_index;
+	unsigned int rx_pring_size;
+	struct efx_tc_flow_rule dflt;
 	struct list_head list;
+	struct list_head rx_list;
+	spinlock_t rx_lock;
+	struct napi_struct napi;
 	struct efx_rep_sw_stats stats;
 };
 
@@ -46,4 +60,10 @@ int efx_ef100_vfrep_create(struct efx_nic *efx, unsigned int i);
 void efx_ef100_vfrep_destroy(struct efx_nic *efx, struct efx_rep *efv);
 void efx_ef100_fini_vfreps(struct efx_nic *efx);
 
+void efx_ef100_rep_rx_packet(struct efx_rep *efv, struct efx_rx_buffer *rx_buf);
+/* Returns the representor corresponding to a VF m-port, or NULL
+ * @mport is an m-port label, *not* an m-port ID!
+ * Caller must hold rcu_read_lock().
+ */
+struct efx_rep *efx_ef100_find_rep_by_mport(struct efx_nic *efx, u16 mport);
 #endif /* EF100_REP_H */

drivers/net/ethernet/sfc/ef100_rx.c

@@ -55,10 +55,14 @@ static bool ef100_has_fcs_error(struct efx_channel *channel, u32 *prefix)
 
 void __ef100_rx_packet(struct efx_channel *channel)
 {
-	struct efx_rx_buffer *rx_buf = efx_rx_buffer(&channel->rx_queue, channel->rx_pkt_index);
+	struct efx_rx_queue *rx_queue = efx_channel_get_rx_queue(channel);
+	struct efx_rx_buffer *rx_buf = efx_rx_buffer(rx_queue,
+						     channel->rx_pkt_index);
 	struct efx_nic *efx = channel->efx;
+	struct ef100_nic_data *nic_data;
 	u8 *eh = efx_rx_buf_va(rx_buf);
 	__wsum csum = 0;
+	u16 ing_port;
 	u32 *prefix;
 
 	prefix = (u32 *)(eh - ESE_GZ_RX_PKT_PREFIX_LEN);
@@ -76,6 +80,37 @@ void __ef100_rx_packet(struct efx_channel *channel)
 		goto out;
 	}
 
+	ing_port = le16_to_cpu((__force __le16) PREFIX_FIELD(prefix, INGRESS_MPORT));
+
+	nic_data = efx->nic_data;
+
+	if (nic_data->have_mport && ing_port != nic_data->base_mport) {
+#ifdef CONFIG_SFC_SRIOV
+		struct efx_rep *efv;
+
+		rcu_read_lock();
+		efv = efx_ef100_find_rep_by_mport(efx, ing_port);
+		if (efv) {
+			if (efv->net_dev->flags & IFF_UP)
+				efx_ef100_rep_rx_packet(efv, rx_buf);
+			rcu_read_unlock();
+			/* Representor Rx doesn't care about PF Rx buffer
+			 * ownership, it just makes a copy. So, we are done
+			 * with the Rx buffer from PF point of view and should
+			 * free it.
+			 */
+			goto free_rx_buffer;
+		}
+		rcu_read_unlock();
+#endif
+		if (net_ratelimit())
+			netif_warn(efx, drv, efx->net_dev,
+				   "Unrecognised ing_port %04x (base %04x), dropping\n",
+				   ing_port, nic_data->base_mport);
+		channel->n_rx_mport_bad++;
+		goto free_rx_buffer;
+	}
+
 	if (likely(efx->net_dev->features & NETIF_F_RXCSUM)) {
 		if (PREFIX_FIELD(prefix, NT_OR_INNER_L3_CLASS) == 1) {
 			++channel->n_rx_ip_hdr_chksum_err;
@@ -87,17 +122,16 @@ void __ef100_rx_packet(struct efx_channel *channel)
 	}
 
 	if (channel->type->receive_skb) {
-		struct efx_rx_queue *rx_queue =
-			efx_channel_get_rx_queue(channel);
-
 		/* no support for special channels yet, so just discard */
 		WARN_ON_ONCE(1);
-		efx_free_rx_buffers(rx_queue, rx_buf, 1);
-		goto out;
+		goto free_rx_buffer;
 	}
 
 	efx_rx_packet_gro(channel, rx_buf, channel->rx_pkt_n_frags, eh, csum);
+	goto out;
 
+free_rx_buffer:
+	efx_free_rx_buffers(rx_queue, rx_buf, 1);
 out:
 	channel->rx_pkt_n_frags = 0;
 }

drivers/net/ethernet/sfc/ef10_sriov.c

@@ -501,14 +501,11 @@ int efx_ef10_sriov_set_vf_mac(struct efx_nic *efx, int vf_i, const u8 *mac)
 		efx_device_detach_sync(vf->efx);
 		efx_net_stop(vf->efx->net_dev);
 
-		down_write(&vf->efx->filter_sem);
 		vf->efx->type->filter_table_remove(vf->efx);
 
 		rc = efx_ef10_vadaptor_free(vf->efx, EVB_PORT_ID_ASSIGNED);
-		if (rc) {
-			up_write(&vf->efx->filter_sem);
+		if (rc)
 			return rc;
-		}
 	}
 
 	rc = efx_ef10_evb_port_assign(efx, EVB_PORT_ID_NULL, vf_i);
@@ -539,12 +536,9 @@ int efx_ef10_sriov_set_vf_mac(struct efx_nic *efx, int vf_i, const u8 *mac)
 	if (vf->efx) {
 		/* VF cannot use the vport_id that the PF created */
 		rc = efx_ef10_vadaptor_alloc(vf->efx, EVB_PORT_ID_ASSIGNED);
-		if (rc) {
-			up_write(&vf->efx->filter_sem);
+		if (rc)
 			return rc;
-		}
 		vf->efx->type->filter_table_probe(vf->efx);
-		up_write(&vf->efx->filter_sem);
 		efx_net_open(vf->efx->net_dev);
 		efx_device_attach_if_not_resetting(vf->efx);
 	}
@@ -580,7 +574,6 @@ int efx_ef10_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i, u16 vlan,
 		efx_net_stop(vf->efx->net_dev);
 
 		mutex_lock(&vf->efx->mac_lock);
-		down_write(&vf->efx->filter_sem);
 		vf->efx->type->filter_table_remove(vf->efx);
 
 		rc = efx_ef10_vadaptor_free(vf->efx, EVB_PORT_ID_ASSIGNED);
@@ -654,7 +647,6 @@ int efx_ef10_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i, u16 vlan,
 		if (rc2)
 			goto reset_nic_up_write;
 
-		up_write(&vf->efx->filter_sem);
 		mutex_unlock(&vf->efx->mac_lock);
 
 		rc2 = efx_net_open(vf->efx->net_dev);
@@ -666,10 +658,8 @@ int efx_ef10_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i, u16 vlan,
 	return rc;
 
 reset_nic_up_write:
-	if (vf->efx) {
-		up_write(&vf->efx->filter_sem);
+	if (vf->efx)
 		mutex_unlock(&vf->efx->mac_lock);
-	}
 reset_nic:
 	if (vf->efx) {
 		netif_err(efx, drv, efx->net_dev,

drivers/net/ethernet/sfc/ethtool_common.c

@@ -91,6 +91,7 @@ static const struct efx_sw_stat_desc efx_sw_stat_desc[] = {
 	EFX_ETHTOOL_UINT_CHANNEL_STAT(rx_xdp_bad_drops),
 	EFX_ETHTOOL_UINT_CHANNEL_STAT(rx_xdp_tx),
 	EFX_ETHTOOL_UINT_CHANNEL_STAT(rx_xdp_redirect),
+	EFX_ETHTOOL_UINT_CHANNEL_STAT(rx_mport_bad),
 #ifdef CONFIG_RFS_ACCEL
 	EFX_ETHTOOL_UINT_CHANNEL_STAT_NO_N(rfs_filter_count),
 	EFX_ETHTOOL_UINT_CHANNEL_STAT(rfs_succeeded),

drivers/net/ethernet/sfc/filter.h

@@ -88,6 +88,7 @@ enum efx_filter_priority {
  *	the automatic filter in its place.
  * @EFX_FILTER_FLAG_RX: Filter is for RX
  * @EFX_FILTER_FLAG_TX: Filter is for TX
+ * @EFX_FILTER_FLAG_VPORT_ID: Virtual port ID for adapter switching.
  */
 enum efx_filter_flags {
 	EFX_FILTER_FLAG_RX_RSS = 0x01,
@@ -95,6 +96,7 @@ enum efx_filter_flags {
 	EFX_FILTER_FLAG_RX_OVER_AUTO = 0x04,
 	EFX_FILTER_FLAG_RX = 0x08,
 	EFX_FILTER_FLAG_TX = 0x10,
+	EFX_FILTER_FLAG_VPORT_ID = 0x20,
 };
 
 /** enum efx_encap_type - types of encapsulation
@@ -127,6 +129,9 @@ enum efx_encap_type {
  *	MCFW context_id.
  * @dmaq_id: Source/target queue index, or %EFX_FILTER_RX_DMAQ_ID_DROP for
  *	an RX drop filter
+ * @vport_id: Virtual port ID associated with RX queue, for adapter switching,
+ *	if %EFX_FILTER_FLAG_VPORT_ID is set.  This is an MCFW vport_id, or on
+ *	EF100 an mport selector.
  * @outer_vid: Outer VLAN ID to match, if %EFX_FILTER_MATCH_OUTER_VID is set
  * @inner_vid: Inner VLAN ID to match, if %EFX_FILTER_MATCH_INNER_VID is set
  * @loc_mac: Local MAC address to match, if %EFX_FILTER_MATCH_LOC_MAC or
@@ -156,6 +161,7 @@ struct efx_filter_spec {
 	u32	priority:2;
 	u32	flags:6;
 	u32	dmaq_id:12;
+	u32	vport_id;
 	u32	rss_context;
 	__be16	outer_vid __aligned(4); /* allow jhash2() of match values */
 	__be16	inner_vid;
@@ -292,6 +298,18 @@ static inline int efx_filter_set_mc_def(struct efx_filter_spec *spec)
 	return 0;
 }
 
+/**
+ * efx_filter_set_vport_id - override virtual port id relating to filter
+ * @spec: Specification to initialise
+ * @vport_id: firmware ID of the virtual port
+ */
+static inline void efx_filter_set_vport_id(struct efx_filter_spec *spec,
+					   u32 vport_id)
+{
+	spec->flags |= EFX_FILTER_FLAG_VPORT_ID;
+	spec->vport_id = vport_id;
+}
+
 static inline void efx_filter_set_encap_type(struct efx_filter_spec *spec,
 					     enum efx_encap_type encap_type)
 {

drivers/net/ethernet/sfc/mae.h

@@ -14,9 +14,29 @@
 /* MCDI interface for the ef100 Match-Action Engine */
 
 #include "net_driver.h"
+#include "tc.h"
+#include "mcdi_pcol.h" /* needed for various MC_CMD_MAE_*_NULL defines */
 
+int efx_mae_allocate_mport(struct efx_nic *efx, u32 *id, u32 *label);
+int efx_mae_free_mport(struct efx_nic *efx, u32 id);
+
+void efx_mae_mport_wire(struct efx_nic *efx, u32 *out);
+void efx_mae_mport_uplink(struct efx_nic *efx, u32 *out);
 void efx_mae_mport_vf(struct efx_nic *efx, u32 vf_id, u32 *out);
+void efx_mae_mport_mport(struct efx_nic *efx, u32 mport_id, u32 *out);
 
 int efx_mae_lookup_mport(struct efx_nic *efx, u32 selector, u32 *id);
 
+int efx_mae_alloc_action_set(struct efx_nic *efx, struct efx_tc_action_set *act);
+int efx_mae_free_action_set(struct efx_nic *efx, u32 fw_id);
+
+int efx_mae_alloc_action_set_list(struct efx_nic *efx,
+				  struct efx_tc_action_set_list *acts);
+int efx_mae_free_action_set_list(struct efx_nic *efx,
+				 struct efx_tc_action_set_list *acts);
+
+int efx_mae_insert_rule(struct efx_nic *efx, const struct efx_tc_match *match,
+			u32 prio, u32 acts_id, u32 *id);
+int efx_mae_delete_rule(struct efx_nic *efx, u32 id);
+
 #endif /* EF100_MAE_H */

drivers/net/ethernet/sfc/mcdi.h

@@ -205,6 +205,8 @@ void efx_mcdi_sensor_event(struct efx_nic *efx, efx_qword_t *ev);
 	((_ofst) + BUILD_BUG_ON_ZERO((_ofst) & (_align - 1)))
 #define _MCDI_DWORD(_buf, _field)					\
 	((_buf) + (_MCDI_CHECK_ALIGN(MC_CMD_ ## _field ## _OFST, 4) >> 2))
+#define _MCDI_STRUCT_DWORD(_buf, _field)				\
+	((_buf) + (_MCDI_CHECK_ALIGN(_field ## _OFST, 4) >> 2))
 
 #define MCDI_BYTE(_buf, _field)						\
 	((void)BUILD_BUG_ON_ZERO(MC_CMD_ ## _field ## _LEN != 1),	\
@@ -214,6 +216,8 @@ void efx_mcdi_sensor_event(struct efx_nic *efx, efx_qword_t *ev);
 	 le16_to_cpu(*(__force const __le16 *)MCDI_PTR(_buf, _field)))
 #define MCDI_SET_DWORD(_buf, _field, _value)				\
 	EFX_POPULATE_DWORD_1(*_MCDI_DWORD(_buf, _field), EFX_DWORD_0, _value)
+#define MCDI_STRUCT_SET_DWORD(_buf, _field, _value)			\
+	EFX_POPULATE_DWORD_1(*_MCDI_STRUCT_DWORD(_buf, _field), EFX_DWORD_0, _value)
 #define MCDI_DWORD(_buf, _field)					\
 	EFX_DWORD_FIELD(*_MCDI_DWORD(_buf, _field), EFX_DWORD_0)
 #define MCDI_POPULATE_DWORD_1(_buf, _field, _name1, _value1)		\

drivers/net/ethernet/sfc/mcdi_filters.c

@@ -221,7 +221,10 @@ static void efx_mcdi_filter_push_prep(struct efx_nic *efx,
 		efx_mcdi_filter_push_prep_set_match_fields(efx, spec, inbuf);
 	}
 
-	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, efx->vport_id);
+	if (flags & EFX_FILTER_FLAG_VPORT_ID)
+		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, spec->vport_id);
+	else
+		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, efx->vport_id);
 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
@@ -488,6 +491,7 @@ static s32 efx_mcdi_filter_insert_locked(struct efx_nic *efx,
 			saved_spec->flags |= spec->flags;
 			saved_spec->rss_context = spec->rss_context;
 			saved_spec->dmaq_id = spec->dmaq_id;
+			saved_spec->vport_id = spec->vport_id;
 		}
 	} else if (!replacing) {
 		kfree(saved_spec);

drivers/net/ethernet/sfc/mcdi_filters.h

@@ -89,6 +89,7 @@ struct efx_mcdi_filter_table {
 	 */
 	bool mc_chaining;
 	bool vlan_filter;
+	/* Entries on the vlan_list are added/removed under filter_sem */
 	struct list_head vlan_list;
 };
 

drivers/net/ethernet/sfc/mcdi_pcol_mae.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+/****************************************************************************
+ * Driver for Solarflare network controllers and boards
+ * Copyright 2019 Solarflare Communications Inc.
+ * Copyright 2019-2022 Xilinx, Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+#ifndef MCDI_PCOL_MAE_H
+#define MCDI_PCOL_MAE_H
+/* MCDI definitions for Match-Action Engine functionality, that are
+ * missing from the main mcdi_pcol.h
+ */
+
+/* MC_CMD_MAE_COUNTER_LIST_ALLOC is not (yet) a released API, but the
+ * following value is needed as an argument to MC_CMD_MAE_ACTION_SET_ALLOC.
+ */
+/* enum: A counter ID that is guaranteed never to represent a real counter */
+#define          MC_CMD_MAE_COUNTER_LIST_ALLOC_OUT_COUNTER_LIST_ID_NULL 0xffffffff
+
+#endif /* MCDI_PCOL_MAE_H */

drivers/net/ethernet/sfc/net_driver.h

@@ -478,6 +478,8 @@ enum efx_sync_events_state {
  * @n_rx_xdp_bad_drops: Count of RX packets dropped due to XDP errors
  * @n_rx_xdp_tx: Count of RX packets retransmitted due to XDP
  * @n_rx_xdp_redirect: Count of RX packets redirected to a different NIC by XDP
+ * @n_rx_mport_bad: Count of RX packets dropped because their ingress mport was
+ *	not recognised
  * @rx_pkt_n_frags: Number of fragments in next packet to be delivered by
  *	__efx_rx_packet(), or zero if there is none
  * @rx_pkt_index: Ring index of first buffer for next packet to be delivered
@@ -540,6 +542,7 @@ struct efx_channel {
 	unsigned int n_rx_xdp_bad_drops;
 	unsigned int n_rx_xdp_tx;
 	unsigned int n_rx_xdp_redirect;
+	unsigned int n_rx_mport_bad;
 
 	unsigned int rx_pkt_n_frags;
 	unsigned int rx_pkt_index;
@@ -975,6 +978,7 @@ enum efx_xdp_tx_queues_mode {
  * @xdp_rxq_info_failed: Have any of the rx queues failed to initialise their
  *      xdp_rxq_info structures?
  * @netdev_notifier: Netdevice notifier.
+ * @tc: state for TC offload (EF100).
  * @mem_bar: The BAR that is mapped into membase.
  * @reg_base: Offset from the start of the bar to the function control window.
  * @monitor_work: Hardware monitor workitem
@@ -1158,6 +1162,7 @@ struct efx_nic {
 	bool xdp_rxq_info_failed;
 
 	struct notifier_block netdev_notifier;
+	struct efx_tc_state *tc;
 
 	unsigned int mem_bar;
 	u32 reg_base;

drivers/net/ethernet/sfc/rx_common.c

@@ -793,7 +793,6 @@ int efx_probe_filters(struct efx_nic *efx)
 	int rc;
 
 	mutex_lock(&efx->mac_lock);
-	down_write(&efx->filter_sem);
 	rc = efx->type->filter_table_probe(efx);
 	if (rc)
 		goto out_unlock;
@@ -830,7 +829,6 @@ int efx_probe_filters(struct efx_nic *efx)
 	}
 #endif
 out_unlock:
-	up_write(&efx->filter_sem);
 	mutex_unlock(&efx->mac_lock);
 	return rc;
 }
@@ -846,9 +844,7 @@ void efx_remove_filters(struct efx_nic *efx)
 		channel->rps_flow_id = NULL;
 	}
 #endif
-	down_write(&efx->filter_sem);
 	efx->type->filter_table_remove(efx);
-	up_write(&efx->filter_sem);
 }
 
 #ifdef CONFIG_RFS_ACCEL

drivers/net/ethernet/sfc/tc.c

+// SPDX-License-Identifier: GPL-2.0-only
+/****************************************************************************
+ * Driver for Solarflare network controllers and boards
+ * Copyright 2019 Solarflare Communications Inc.
+ * Copyright 2020-2022 Xilinx Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+#include "tc.h"
+#include "mae.h"
+#include "ef100_rep.h"
+#include "efx.h"
+
+static void efx_tc_free_action_set(struct efx_nic *efx,
+				   struct efx_tc_action_set *act, bool in_hw)
+{
+	/* Failure paths calling this on the 'running action' set in_hw=false,
+	 * because if the alloc had succeeded we'd've put it in acts.list and
+	 * not still have it in act.
+	 */
+	if (in_hw) {
+		efx_mae_free_action_set(efx, act->fw_id);
+		/* in_hw is true iff we are on an acts.list; make sure to
+		 * remove ourselves from that list before we are freed.
+		 */
+		list_del(&act->list);
+	}
+	kfree(act);
+}
+
+static void efx_tc_free_action_set_list(struct efx_nic *efx,
+					struct efx_tc_action_set_list *acts,
+					bool in_hw)
+{
+	struct efx_tc_action_set *act, *next;
+
+	/* Failure paths set in_hw=false, because usually the acts didn't get
+	 * to efx_mae_alloc_action_set_list(); if they did, the failure tree
+	 * has a separate efx_mae_free_action_set_list() before calling us.
+	 */
+	if (in_hw)
+		efx_mae_free_action_set_list(efx, acts);
+	/* Any act that's on the list will be in_hw even if the list isn't */
+	list_for_each_entry_safe(act, next, &acts->list, list)
+		efx_tc_free_action_set(efx, act, true);
+	/* Don't kfree, as acts is embedded inside a struct efx_tc_flow_rule */
+}
+
+static void efx_tc_delete_rule(struct efx_nic *efx, struct efx_tc_flow_rule *rule)
+{
+	efx_mae_delete_rule(efx, rule->fw_id);
+
+	/* Release entries in subsidiary tables */
+	efx_tc_free_action_set_list(efx, &rule->acts, true);
+	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+}
+
+static int efx_tc_configure_default_rule(struct efx_nic *efx, u32 ing_port,
+					 u32 eg_port, struct efx_tc_flow_rule *rule)
+{
+	struct efx_tc_action_set_list *acts = &rule->acts;
+	struct efx_tc_match *match = &rule->match;
+	struct efx_tc_action_set *act;
+	int rc;
+
+	match->value.ingress_port = ing_port;
+	match->mask.ingress_port = ~0;
+	act = kzalloc(sizeof(*act), GFP_KERNEL);
+	if (!act)
+		return -ENOMEM;
+	act->deliver = 1;
+	act->dest_mport = eg_port;
+	rc = efx_mae_alloc_action_set(efx, act);
+	if (rc)
+		goto fail1;
+	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
+	list_add_tail(&act->list, &acts->list);
+	rc = efx_mae_alloc_action_set_list(efx, acts);
+	if (rc)
+		goto fail2;
+	rc = efx_mae_insert_rule(efx, match, EFX_TC_PRIO_DFLT,
+				 acts->fw_id, &rule->fw_id);
+	if (rc)
+		goto fail3;
+	return 0;
+fail3:
+	efx_mae_free_action_set_list(efx, acts);
+fail2:
+	list_del(&act->list);
+	efx_mae_free_action_set(efx, act->fw_id);
+fail1:
+	kfree(act);
+	return rc;
+}
+
+static int efx_tc_configure_default_rule_pf(struct efx_nic *efx)
+{
+	struct efx_tc_flow_rule *rule = &efx->tc->dflt.pf;
+	u32 ing_port, eg_port;
+
+	efx_mae_mport_uplink(efx, &ing_port);
+	efx_mae_mport_wire(efx, &eg_port);
+	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
+}
+
+static int efx_tc_configure_default_rule_wire(struct efx_nic *efx)
+{
+	struct efx_tc_flow_rule *rule = &efx->tc->dflt.wire;
+	u32 ing_port, eg_port;
+
+	efx_mae_mport_wire(efx, &ing_port);
+	efx_mae_mport_uplink(efx, &eg_port);
+	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
+}
+
+int efx_tc_configure_default_rule_rep(struct efx_rep *efv)
+{
+	struct efx_tc_flow_rule *rule = &efv->dflt;
+	struct efx_nic *efx = efv->parent;
+	u32 ing_port, eg_port;
+
+	efx_mae_mport_mport(efx, efv->mport, &ing_port);
+	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
+	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
+}
+
+void efx_tc_deconfigure_default_rule(struct efx_nic *efx,
+				     struct efx_tc_flow_rule *rule)
+{
+	if (rule->fw_id != MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL)
+		efx_tc_delete_rule(efx, rule);
+	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+}
+
+static int efx_tc_configure_rep_mport(struct efx_nic *efx)
+{
+	u32 rep_mport_label;
+	int rc;
+
+	rc = efx_mae_allocate_mport(efx, &efx->tc->reps_mport_id, &rep_mport_label);
+	if (rc)
+		return rc;
+	pci_dbg(efx->pci_dev, "created rep mport 0x%08x (0x%04x)\n",
+		efx->tc->reps_mport_id, rep_mport_label);
+	/* Use mport *selector* as vport ID */
+	efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
+			    &efx->tc->reps_mport_vport_id);
+	return 0;
+}
+
+static void efx_tc_deconfigure_rep_mport(struct efx_nic *efx)
+{
+	efx_mae_free_mport(efx, efx->tc->reps_mport_id);
+	efx->tc->reps_mport_id = MAE_MPORT_SELECTOR_NULL;
+}
+
+int efx_tc_insert_rep_filters(struct efx_nic *efx)
+{
+	struct efx_filter_spec promisc, allmulti;
+	int rc;
+
+	if (efx->type->is_vf)
+		return 0;
+	if (!efx->tc)
+		return 0;
+	efx_filter_init_rx(&promisc, EFX_FILTER_PRI_REQUIRED, 0, 0);
+	efx_filter_set_uc_def(&promisc);
+	efx_filter_set_vport_id(&promisc, efx->tc->reps_mport_vport_id);
+	rc = efx_filter_insert_filter(efx, &promisc, false);
+	if (rc < 0)
+		return rc;
+	efx->tc->reps_filter_uc = rc;
+	efx_filter_init_rx(&allmulti, EFX_FILTER_PRI_REQUIRED, 0, 0);
+	efx_filter_set_mc_def(&allmulti);
+	efx_filter_set_vport_id(&allmulti, efx->tc->reps_mport_vport_id);
+	rc = efx_filter_insert_filter(efx, &allmulti, false);
+	if (rc < 0)
+		return rc;
+	efx->tc->reps_filter_mc = rc;
+	return 0;
+}
+
+void efx_tc_remove_rep_filters(struct efx_nic *efx)
+{
+	if (efx->type->is_vf)
+		return;
+	if (!efx->tc)
+		return;
+	if (efx->tc->reps_filter_mc >= 0)
+		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_mc);
+	efx->tc->reps_filter_mc = -1;
+	if (efx->tc->reps_filter_uc >= 0)
+		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_uc);
+	efx->tc->reps_filter_uc = -1;
+}
+
+int efx_init_tc(struct efx_nic *efx)
+{
+	int rc;
+
+	rc = efx_tc_configure_default_rule_pf(efx);
+	if (rc)
+		return rc;
+	rc = efx_tc_configure_default_rule_wire(efx);
+	if (rc)
+		return rc;
+	return efx_tc_configure_rep_mport(efx);
+}
+
+void efx_fini_tc(struct efx_nic *efx)
+{
+	/* We can get called even if efx_init_struct_tc() failed */
+	if (!efx->tc)
+		return;
+	efx_tc_deconfigure_rep_mport(efx);
+	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.pf);
+	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.wire);
+}
+
+int efx_init_struct_tc(struct efx_nic *efx)
+{
+	if (efx->type->is_vf)
+		return 0;
+
+	efx->tc = kzalloc(sizeof(*efx->tc), GFP_KERNEL);
+	if (!efx->tc)
+		return -ENOMEM;
+
+	efx->tc->reps_filter_uc = -1;
+	efx->tc->reps_filter_mc = -1;
+	INIT_LIST_HEAD(&efx->tc->dflt.pf.acts.list);
+	efx->tc->dflt.pf.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+	INIT_LIST_HEAD(&efx->tc->dflt.wire.acts.list);
+	efx->tc->dflt.wire.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+	return 0;
+}
+
+void efx_fini_struct_tc(struct efx_nic *efx)
+{
+	if (!efx->tc)
+		return;
+
+	EFX_WARN_ON_PARANOID(efx->tc->dflt.pf.fw_id !=
+			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
+	EFX_WARN_ON_PARANOID(efx->tc->dflt.wire.fw_id !=
+			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
+	kfree(efx->tc);
+	efx->tc = NULL;
+}

drivers/net/ethernet/sfc/tc.h

+/* SPDX-License-Identifier: GPL-2.0-only */
+/****************************************************************************
+ * Driver for Solarflare network controllers and boards
+ * Copyright 2019 Solarflare Communications Inc.
+ * Copyright 2020-2022 Xilinx Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation, incorporated herein by reference.
+ */
+
+#ifndef EFX_TC_H
+#define EFX_TC_H
+#include "net_driver.h"
+
+struct efx_tc_action_set {
+	u16 deliver:1;
+	u32 dest_mport;
+	u32 fw_id; /* index of this entry in firmware actions table */
+	struct list_head list;
+};
+
+struct efx_tc_match_fields {
+	/* L1 */
+	u32 ingress_port;
+};
+
+struct efx_tc_match {
+	struct efx_tc_match_fields value;
+	struct efx_tc_match_fields mask;
+};
+
+struct efx_tc_action_set_list {
+	struct list_head list;
+	u32 fw_id;
+};
+
+struct efx_tc_flow_rule {
+	struct efx_tc_match match;
+	struct efx_tc_action_set_list acts;
+	u32 fw_id;
+};
+
+enum efx_tc_rule_prios {
+	EFX_TC_PRIO_DFLT, /* Default switch rule; one of efx_tc_default_rules */
+	EFX_TC_PRIO__NUM
+};
+
+/**
+ * struct efx_tc_state - control plane data for TC offload
+ *
+ * @reps_mport_id: MAE port allocated for representor RX
+ * @reps_filter_uc: VNIC filter for representor unicast RX (promisc)
+ * @reps_filter_mc: VNIC filter for representor multicast RX (allmulti)
+ * @reps_mport_vport_id: vport_id for representor RX filters
+ * @dflt: Match-action rules for default switching; at priority
+ *	%EFX_TC_PRIO_DFLT.  Named by *ingress* port
+ * @dflt.pf: rule for traffic ingressing from PF (egresses to wire)
+ * @dflt.wire: rule for traffic ingressing from wire (egresses to PF)
+ */
+struct efx_tc_state {
+	u32 reps_mport_id, reps_mport_vport_id;
+	s32 reps_filter_uc, reps_filter_mc;
+	struct {
+		struct efx_tc_flow_rule pf;
+		struct efx_tc_flow_rule wire;
+	} dflt;
+};
+
+struct efx_rep;
+
+int efx_tc_configure_default_rule_rep(struct efx_rep *efv);
+void efx_tc_deconfigure_default_rule(struct efx_nic *efx,
+				     struct efx_tc_flow_rule *rule);
+
+int efx_tc_insert_rep_filters(struct efx_nic *efx);
+void efx_tc_remove_rep_filters(struct efx_nic *efx);
+
+int efx_init_tc(struct efx_nic *efx);
+void efx_fini_tc(struct efx_nic *efx);
+
+int efx_init_struct_tc(struct efx_nic *efx);
+void efx_fini_struct_tc(struct efx_nic *efx);
+
+#endif /* EFX_TC_H */