sfc: support TC decap rules

index 2b07bb2fd735..34c1ff87ba5e 100644
@@ -66,7 +66,7 @@ static const struct rhashtable_params efx_tc_match_action_ht_params = {
 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,
+	/* Failure paths calling this on the 'cursor' 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.
 	 */

@@ -407,6 +407,30 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
 		goto release;
 	}
 
+	/**
+	 * DOC: TC action translation
+	 *
+	 * Actions in TC are sequential and cumulative, with delivery actions
+	 * potentially anywhere in the order.  The EF100 MAE, however, takes
+	 * an 'action set list' consisting of 'action sets', each of which is
+	 * applied to the _original_ packet, and consists of a set of optional
+	 * actions in a fixed order with delivery at the end.
+	 * To translate between these two models, we maintain a 'cursor', @act,
+	 * which describes the cumulative effect of all the packet-mutating
+	 * actions encountered so far; on handling a delivery (mirred or drop)
+	 * action, once the action-set has been inserted into hardware, we
+	 * append @act to the action-set list (@rule->acts); if this is a pipe
+	 * action (mirred mirror) we then allocate a new @act with a copy of
+	 * the cursor state _before_ the delivery action, otherwise we set @act
+	 * to %NULL.
+	 * This ensures that every allocated action-set is either attached to
+	 * @rule->acts or pointed to by @act (and never both), and that only
+	 * those action-sets in @rule->acts exist in hardware.  Consequently,
+	 * in the failure path, @act only needs to be freed in memory, whereas
+	 * for @rule->acts we remove each action-set from hardware before
+	 * freeing it (efx_tc_free_action_set_list()), even if the action-set
+	 * list itself is not in hardware.
+	 */
 	flow_action_for_each(i, fa, &fr->action) {
 		struct efx_tc_action_set save;
 		u16 tci;

index c53d354c1fb2..2290a63908c5 100644
@@ -254,13 +254,23 @@ static int efx_mae_get_rule_fields(struct efx_nic *efx, u32 cmd,
 	size_t outlen;
 	int rc, i;
 
+	/* AR and OR caps MCDIs have identical layout, so we are using the
+	 * same code for both.
+	 */
+	BUILD_BUG_ON(MC_CMD_MAE_GET_AR_CAPS_OUT_LEN(MAE_NUM_FIELDS) <
+		     MC_CMD_MAE_GET_OR_CAPS_OUT_LEN(MAE_NUM_FIELDS));
 	BUILD_BUG_ON(MC_CMD_MAE_GET_AR_CAPS_IN_LEN);
+	BUILD_BUG_ON(MC_CMD_MAE_GET_OR_CAPS_IN_LEN);
 
 	rc = efx_mcdi_rpc(efx, cmd, NULL, 0, outbuf, sizeof(outbuf), &outlen);
 	if (rc)
 		return rc;
+	BUILD_BUG_ON(MC_CMD_MAE_GET_AR_CAPS_OUT_COUNT_OFST !=
+		     MC_CMD_MAE_GET_OR_CAPS_OUT_COUNT_OFST);
 	count = MCDI_DWORD(outbuf, MAE_GET_AR_CAPS_OUT_COUNT);
 	memset(field_support, MAE_FIELD_UNSUPPORTED, MAE_NUM_FIELDS);
+	BUILD_BUG_ON(MC_CMD_MAE_GET_AR_CAPS_OUT_FIELD_FLAGS_OFST !=
+		     MC_CMD_MAE_GET_OR_CAPS_OUT_FIELD_FLAGS_OFST);
 	caps = _MCDI_DWORD(outbuf, MAE_GET_AR_CAPS_OUT_FIELD_FLAGS);
 	/* We're only interested in the support status enum, not any other
 	 * flags, so just extract that from each entry.

@@ -278,8 +288,12 @@ int efx_mae_get_caps(struct efx_nic *efx, struct mae_caps *caps)
 	rc = efx_mae_get_basic_caps(efx, caps);
 	if (rc)
 		return rc;
-	return efx_mae_get_rule_fields(efx, MC_CMD_MAE_GET_AR_CAPS,
-				       caps->action_rule_fields);
+	rc = efx_mae_get_rule_fields(efx, MC_CMD_MAE_GET_AR_CAPS,
+				     caps->action_rule_fields);
+	if (rc)
+		return rc;
+	return efx_mae_get_rule_fields(efx, MC_CMD_MAE_GET_OR_CAPS,
+				       caps->outer_rule_fields);
 }
 
 /* Bit twiddling:

@@ -432,11 +446,67 @@ int efx_mae_match_check_caps(struct efx_nic *efx,
 	    CHECK_BIT(IP_FIRST_FRAG, ip_firstfrag) ||
 	    CHECK(RECIRC_ID, recirc_id))
 		return rc;
+	/* Matches on outer fields are done in a separate hardware table,
+	 * the Outer Rule table.  Thus the Action Rule merely does an
+	 * exact match on Outer Rule ID if any outer field matches are
+	 * present.  The exception is the VNI/VSID (enc_keyid), which is
+	 * available to the Action Rule match iff the Outer Rule matched
+	 * (and thus identified the encap protocol to use to extract it).
+	 */
+	if (efx_tc_match_is_encap(mask)) {
+		rc = efx_mae_match_check_cap_typ(
+				supported_fields[MAE_FIELD_OUTER_RULE_ID],
+				MASK_ONES);
+		if (rc) {
+			NL_SET_ERR_MSG_MOD(extack, "No support for encap rule ID matches");
+			return rc;
+		}
+		if (CHECK(ENC_VNET_ID, enc_keyid))
+			return rc;
+	} else if (mask->enc_keyid) {
+		NL_SET_ERR_MSG_MOD(extack, "Match on enc_keyid requires other encap fields");
+		return -EINVAL;
+	}
 	return 0;
 }
 #undef CHECK_BIT
 #undef CHECK
 
+#define CHECK(_mcdi)	({						       \
+	rc = efx_mae_match_check_cap_typ(supported_fields[MAE_FIELD_ ## _mcdi],\
+					 MASK_ONES);			       \
+	if (rc)								       \
+		NL_SET_ERR_MSG_FMT_MOD(extack,				       \
+				       "No support for field %s", #_mcdi);     \
+	rc;								       \
+})
+/* Checks that the fields needed for encap-rule matches are supported by the
+ * MAE.  All the fields are exact-match.
+ */
+int efx_mae_check_encap_match_caps(struct efx_nic *efx, bool ipv6,
+				   struct netlink_ext_ack *extack)
+{
+	u8 *supported_fields = efx->tc->caps->outer_rule_fields;
+	int rc;
+
+	if (CHECK(ENC_ETHER_TYPE))
+		return rc;
+	if (ipv6) {
+		if (CHECK(ENC_SRC_IP6) ||
+		    CHECK(ENC_DST_IP6))
+			return rc;
+	} else {
+		if (CHECK(ENC_SRC_IP4) ||
+		    CHECK(ENC_DST_IP4))
+			return rc;
+	}
+	if (CHECK(ENC_L4_DPORT) ||
+	    CHECK(ENC_IP_PROTO))
+		return rc;
+	return 0;
+}
+#undef CHECK
+
 int efx_mae_allocate_counter(struct efx_nic *efx, struct efx_tc_counter *cnt)
 {
 	MCDI_DECLARE_BUF(outbuf, MC_CMD_MAE_COUNTER_ALLOC_OUT_LEN(1));

@@ -941,6 +1011,29 @@ static int efx_mae_populate_match_criteria(MCDI_DECLARE_STRUCT_PTR(match_crit),
 				match->value.tcp_flags);
 	MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_FIELD_MASK_VALUE_PAIRS_V2_TCP_FLAGS_BE_MASK,
 				match->mask.tcp_flags);
+	/* enc-keys are handled indirectly, through encap_match ID */
+	if (match->encap) {
+		MCDI_STRUCT_SET_DWORD(match_crit, MAE_FIELD_MASK_VALUE_PAIRS_V2_OUTER_RULE_ID,
+				      match->encap->fw_id);
+		MCDI_STRUCT_SET_DWORD(match_crit, MAE_FIELD_MASK_VALUE_PAIRS_V2_OUTER_RULE_ID_MASK,
+				      U32_MAX);
+		/* enc_keyid (VNI/VSID) is not part of the encap_match */
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_FIELD_MASK_VALUE_PAIRS_V2_ENC_VNET_ID_BE,
+					 match->value.enc_keyid);
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_FIELD_MASK_VALUE_PAIRS_V2_ENC_VNET_ID_BE_MASK,
+					 match->mask.enc_keyid);
+	} else if (WARN_ON_ONCE(match->mask.enc_src_ip) ||
+		   WARN_ON_ONCE(match->mask.enc_dst_ip) ||
+		   WARN_ON_ONCE(!ipv6_addr_any(&match->mask.enc_src_ip6)) ||
+		   WARN_ON_ONCE(!ipv6_addr_any(&match->mask.enc_dst_ip6)) ||
+		   WARN_ON_ONCE(match->mask.enc_ip_tos) ||
+		   WARN_ON_ONCE(match->mask.enc_ip_ttl) ||
+		   WARN_ON_ONCE(match->mask.enc_sport) ||
+		   WARN_ON_ONCE(match->mask.enc_dport) ||
+		   WARN_ON_ONCE(match->mask.enc_keyid)) {
+		/* No enc-keys should appear in a rule without an encap_match */
+		return -EOPNOTSUPP;
+	}
 	return 0;
 }
 

index bec293a06733..2ccbc62d79b9 100644
@@ -72,6 +72,7 @@ struct mae_caps {
 	u32 match_field_count;
 	u32 action_prios;
 	u8 action_rule_fields[MAE_NUM_FIELDS];
+	u8 outer_rule_fields[MAE_NUM_FIELDS];
 };
 
 int efx_mae_get_caps(struct efx_nic *efx, struct mae_caps *caps);

@@ -79,6 +80,8 @@ int efx_mae_get_caps(struct efx_nic *efx, struct mae_caps *caps);
 int efx_mae_match_check_caps(struct efx_nic *efx,
 			     const struct efx_tc_match_fields *mask,
 			     struct netlink_ext_ack *extack);
+int efx_mae_check_encap_match_caps(struct efx_nic *efx, bool ipv6,
+				   struct netlink_ext_ack *extack);
 
 int efx_mae_allocate_counter(struct efx_nic *efx, struct efx_tc_counter *cnt);
 int efx_mae_free_counter(struct efx_nic *efx, struct efx_tc_counter *cnt);

index 542853f60c2a..c1485679507c 100644
@@ -48,11 +48,35 @@ struct efx_tc_match_fields {
 	/* L4 */
 	__be16 l4_sport, l4_dport; /* Ports (UDP, TCP) */
 	__be16 tcp_flags;
+	/* Encap.  The following are *outer* fields.  Note that there are no
+	 * outer eth (L2) fields; this is because TC doesn't have them.
+	 */
+	__be32 enc_src_ip, enc_dst_ip;
+	struct in6_addr enc_src_ip6, enc_dst_ip6;
+	u8 enc_ip_tos, enc_ip_ttl;
+	__be16 enc_sport, enc_dport;
+	__be32 enc_keyid; /* e.g. VNI, VSID */
+};
+
+static inline bool efx_tc_match_is_encap(const struct efx_tc_match_fields *mask)
+{
+	return mask->enc_src_ip || mask->enc_dst_ip ||
+	       !ipv6_addr_any(&mask->enc_src_ip6) ||
+	       !ipv6_addr_any(&mask->enc_dst_ip6) || mask->enc_ip_tos ||
+	       mask->enc_ip_ttl || mask->enc_sport || mask->enc_dport;
+}
+
+struct efx_tc_encap_match {
+	__be32 src_ip, dst_ip;
+	struct in6_addr src_ip6, dst_ip6;
+	__be16 udp_dport;
+	u32 fw_id; /* index of this entry in firmware encap match table */
 };
 
 struct efx_tc_match {
 	struct efx_tc_match_fields value;
 	struct efx_tc_match_fields mask;
+	struct efx_tc_encap_match *encap;
 };
 
 struct efx_tc_action_set_list {

index 34c1ff87ba5e..21eb79b20978 100644
@@ -193,6 +193,11 @@ static int efx_tc_flower_parse_match(struct efx_nic *efx,
 	      BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
 	      BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
 	      BIT(FLOW_DISSECTOR_KEY_PORTS) |
+	      BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
+	      BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
+	      BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
+	      BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
+	      BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
 	      BIT(FLOW_DISSECTOR_KEY_TCP) |
 	      BIT(FLOW_DISSECTOR_KEY_IP))) {
 		NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported flower keys %#x",

@@ -280,6 +285,57 @@ static int efx_tc_flower_parse_match(struct efx_nic *efx,
 	MAP_KEY_AND_MASK(PORTS, ports, src, l4_sport);
 	MAP_KEY_AND_MASK(PORTS, ports, dst, l4_dport);
 	MAP_KEY_AND_MASK(TCP, tcp, flags, tcp_flags);
+	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
+		struct flow_match_control fm;
+
+		flow_rule_match_enc_control(rule, &fm);
+		if (fm.mask->flags) {
+			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported match on enc_control.flags %#x",
+					       fm.mask->flags);
+			return -EOPNOTSUPP;
+		}
+		if (!IS_ALL_ONES(fm.mask->addr_type)) {
+			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported enc addr_type mask %u (key %u)",
+					       fm.mask->addr_type,
+					       fm.key->addr_type);
+			return -EOPNOTSUPP;
+		}
+		switch (fm.key->addr_type) {
+		case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
+			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
+					     src, enc_src_ip);
+			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
+					     dst, enc_dst_ip);
+			break;
+#ifdef CONFIG_IPV6
+		case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
+			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
+					     src, enc_src_ip6);
+			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
+					     dst, enc_dst_ip6);
+			break;
+#endif
+		default:
+			NL_SET_ERR_MSG_FMT_MOD(extack,
+					       "Unsupported enc addr_type %u (supported are IPv4, IPv6)",
+					       fm.key->addr_type);
+			return -EOPNOTSUPP;
+		}
+		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, tos, enc_ip_tos);
+		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, ttl, enc_ip_ttl);
+		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, src, enc_sport);
+		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, dst, enc_dport);
+		MAP_ENC_KEY_AND_MASK(KEYID, enc_keyid, enc_keyid, keyid, enc_keyid);
+	} else if (dissector->used_keys &
+		   (BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
+		    BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
+		    BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
+		    BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
+		    BIT(FLOW_DISSECTOR_KEY_ENC_PORTS))) {
+		NL_SET_ERR_MSG_FMT_MOD(extack, "Flower enc keys require enc_control (keys: %#x)",
+				       dissector->used_keys);
+		return -EOPNOTSUPP;
+	}
 
 	return 0;
 }

@@ -373,6 +429,11 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
 	rc = efx_tc_flower_parse_match(efx, fr, &match, extack);
 	if (rc)
 		return rc;
+	if (efx_tc_match_is_encap(&match.mask)) {
+		NL_SET_ERR_MSG_MOD(extack, "Ingress enc_key matches not supported");
+		rc = -EOPNOTSUPP;
+		goto release;
+	}
 
 	if (tc->common.chain_index) {
 		NL_SET_ERR_MSG_MOD(extack, "No support for nonzero chain_index");

index 2290a63908c5..92f1383ee4b9 100644
@@ -558,6 +558,20 @@ int efx_mae_free_counter(struct efx_nic *efx, struct efx_tc_counter *cnt)
 	return 0;
 }
 
+static int efx_mae_encap_type_to_mae_type(enum efx_encap_type type)
+{
+	switch (type & EFX_ENCAP_TYPES_MASK) {
+	case EFX_ENCAP_TYPE_NONE:
+		return MAE_MCDI_ENCAP_TYPE_NONE;
+	case EFX_ENCAP_TYPE_VXLAN:
+		return MAE_MCDI_ENCAP_TYPE_VXLAN;
+	case EFX_ENCAP_TYPE_GENEVE:
+		return MAE_MCDI_ENCAP_TYPE_GENEVE;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
 int efx_mae_lookup_mport(struct efx_nic *efx, u32 vf_idx, u32 *id)
 {
 	struct ef100_nic_data *nic_data = efx->nic_data;

@@ -915,6 +929,97 @@ int efx_mae_free_action_set_list(struct efx_nic *efx,
 	return 0;
 }
 
+int efx_mae_register_encap_match(struct efx_nic *efx,
+				 struct efx_tc_encap_match *encap)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAE_OUTER_RULE_INSERT_IN_LEN(MAE_ENC_FIELD_PAIRS_LEN));
+	MCDI_DECLARE_BUF(outbuf, MC_CMD_MAE_OUTER_RULE_INSERT_OUT_LEN);
+	MCDI_DECLARE_STRUCT_PTR(match_crit);
+	size_t outlen;
+	int rc;
+
+	rc = efx_mae_encap_type_to_mae_type(encap->tun_type);
+	if (rc < 0)
+		return rc;
+	match_crit = _MCDI_DWORD(inbuf, MAE_OUTER_RULE_INSERT_IN_FIELD_MATCH_CRITERIA);
+	/* The struct contains IP src and dst, and udp dport.
+	 * So we actually need to filter on IP src and dst, L4 dport, and
+	 * ipproto == udp.
+	 */
+	MCDI_SET_DWORD(inbuf, MAE_OUTER_RULE_INSERT_IN_ENCAP_TYPE, rc);
+#ifdef CONFIG_IPV6
+	if (encap->src_ip | encap->dst_ip) {
+#endif
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_SRC_IP4_BE,
+					 encap->src_ip);
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_SRC_IP4_BE_MASK,
+					 ~(__be32)0);
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_DST_IP4_BE,
+					 encap->dst_ip);
+		MCDI_STRUCT_SET_DWORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_DST_IP4_BE_MASK,
+					 ~(__be32)0);
+		MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_ETHER_TYPE_BE,
+					htons(ETH_P_IP));
+#ifdef CONFIG_IPV6
+	} else {
+		memcpy(MCDI_STRUCT_PTR(match_crit, MAE_ENC_FIELD_PAIRS_ENC_SRC_IP6_BE),
+		       &encap->src_ip6, sizeof(encap->src_ip6));
+		memset(MCDI_STRUCT_PTR(match_crit, MAE_ENC_FIELD_PAIRS_ENC_SRC_IP6_BE_MASK),
+		       0xff, sizeof(encap->src_ip6));
+		memcpy(MCDI_STRUCT_PTR(match_crit, MAE_ENC_FIELD_PAIRS_ENC_DST_IP6_BE),
+		       &encap->dst_ip6, sizeof(encap->dst_ip6));
+		memset(MCDI_STRUCT_PTR(match_crit, MAE_ENC_FIELD_PAIRS_ENC_DST_IP6_BE_MASK),
+		       0xff, sizeof(encap->dst_ip6));
+		MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_ETHER_TYPE_BE,
+					htons(ETH_P_IPV6));
+	}
+#endif
+	MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_ETHER_TYPE_BE_MASK,
+				~(__be16)0);
+	MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_L4_DPORT_BE,
+				encap->udp_dport);
+	MCDI_STRUCT_SET_WORD_BE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_L4_DPORT_BE_MASK,
+				~(__be16)0);
+	MCDI_STRUCT_SET_BYTE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_IP_PROTO, IPPROTO_UDP);
+	MCDI_STRUCT_SET_BYTE(match_crit, MAE_ENC_FIELD_PAIRS_ENC_IP_PROTO_MASK, ~0);
+	rc = efx_mcdi_rpc(efx, MC_CMD_MAE_OUTER_RULE_INSERT, inbuf,
+			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
+	if (rc)
+		return rc;
+	if (outlen < sizeof(outbuf))
+		return -EIO;
+	encap->fw_id = MCDI_DWORD(outbuf, MAE_OUTER_RULE_INSERT_OUT_OR_ID);
+	return 0;
+}
+
+int efx_mae_unregister_encap_match(struct efx_nic *efx,
+				   struct efx_tc_encap_match *encap)
+{
+	MCDI_DECLARE_BUF(outbuf, MC_CMD_MAE_OUTER_RULE_REMOVE_OUT_LEN(1));
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAE_OUTER_RULE_REMOVE_IN_LEN(1));
+	size_t outlen;
+	int rc;
+
+	MCDI_SET_DWORD(inbuf, MAE_OUTER_RULE_REMOVE_IN_OR_ID, encap->fw_id);
+	rc = efx_mcdi_rpc(efx, MC_CMD_MAE_OUTER_RULE_REMOVE, inbuf,
+			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
+	if (rc)
+		return rc;
+	if (outlen < sizeof(outbuf))
+		return -EIO;
+	/* FW freed a different ID than we asked for, should also never happen.
+	 * Warn because it means we've now got a different idea to the FW of
+	 * what encap_mds exist, which could cause mayhem later.
+	 */
+	if (WARN_ON(MCDI_DWORD(outbuf, MAE_OUTER_RULE_REMOVE_OUT_REMOVED_OR_ID) != encap->fw_id))
+		return -EIO;
+	/* We're probably about to free @encap, but let's just make sure its
+	 * fw_id is blatted so that it won't look valid if it leaks out.
+	 */
+	encap->fw_id = MC_CMD_MAE_OUTER_RULE_INSERT_OUT_OUTER_RULE_ID_NULL;
+	return 0;
+}
+
 static int efx_mae_populate_match_criteria(MCDI_DECLARE_STRUCT_PTR(match_crit),
 					   const struct efx_tc_match *match)
 {

index 2ccbc62d79b9..79d315a129ab 100644
@@ -94,6 +94,11 @@ int efx_mae_alloc_action_set_list(struct efx_nic *efx,
 int efx_mae_free_action_set_list(struct efx_nic *efx,
 				 struct efx_tc_action_set_list *acts);
 
+int efx_mae_register_encap_match(struct efx_nic *efx,
+				 struct efx_tc_encap_match *encap);
+int efx_mae_unregister_encap_match(struct efx_nic *efx,
+				   struct efx_tc_encap_match *encap);
+
 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);

index c1485679507c..19782c9a4354 100644
@@ -70,6 +70,7 @@ struct efx_tc_encap_match {
 	__be32 src_ip, dst_ip;
 	struct in6_addr src_ip6, dst_ip6;
 	__be16 udp_dport;
+	u16 tun_type; /* enum efx_encap_type */
 	u32 fw_id; /* index of this entry in firmware encap match table */
 };
 

index 21eb79b20978..f18ac2a7697a 100644
@@ -57,6 +57,12 @@ static s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv
 	return mport;
 }
 
+static const struct rhashtable_params efx_tc_encap_match_ht_params = {
+	.key_len	= offsetof(struct efx_tc_encap_match, linkage),
+	.key_offset	= 0,
+	.head_offset	= offsetof(struct efx_tc_encap_match, linkage),
+};
+
 static const struct rhashtable_params efx_tc_match_action_ht_params = {
 	.key_len	= sizeof(unsigned long),
 	.key_offset	= offsetof(struct efx_tc_flow_rule, cookie),

@@ -340,6 +346,144 @@ static int efx_tc_flower_parse_match(struct efx_nic *efx,
 	return 0;
 }
 
+__always_unused
+static int efx_tc_flower_record_encap_match(struct efx_nic *efx,
+					    struct efx_tc_match *match,
+					    enum efx_encap_type type,
+					    struct netlink_ext_ack *extack)
+{
+	struct efx_tc_encap_match *encap, *old;
+	bool ipv6;
+	int rc;
+
+	/* We require that the socket-defining fields (IP addrs and UDP dest
+	 * port) are present and exact-match.  Other fields are currently not
+	 * allowed.  This meets what OVS will ask for, and means that we don't
+	 * need to handle difficult checks for overlapping matches as could
+	 * come up if we allowed masks or varying sets of match fields.
+	 */
+	if (match->mask.enc_dst_ip | match->mask.enc_src_ip) {
+		if (!IS_ALL_ONES(match->mask.enc_dst_ip)) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match is not exact on dst IP address");
+			return -EOPNOTSUPP;
+		}
+		if (!IS_ALL_ONES(match->mask.enc_src_ip)) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match is not exact on src IP address");
+			return -EOPNOTSUPP;
+		}
+#ifdef CONFIG_IPV6
+		if (!ipv6_addr_any(&match->mask.enc_dst_ip6) ||
+		    !ipv6_addr_any(&match->mask.enc_src_ip6)) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match on both IPv4 and IPv6, don't understand");
+			return -EOPNOTSUPP;
+		}
+	} else {
+		ipv6 = true;
+		if (!efx_ipv6_addr_all_ones(&match->mask.enc_dst_ip6)) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match is not exact on dst IP address");
+			return -EOPNOTSUPP;
+		}
+		if (!efx_ipv6_addr_all_ones(&match->mask.enc_src_ip6)) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match is not exact on src IP address");
+			return -EOPNOTSUPP;
+		}
+#endif
+	}
+	if (!IS_ALL_ONES(match->mask.enc_dport)) {
+		NL_SET_ERR_MSG_MOD(extack, "Egress encap match is not exact on dst UDP port");
+		return -EOPNOTSUPP;
+	}
+	if (match->mask.enc_sport) {
+		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on src UDP port not supported");
+		return -EOPNOTSUPP;
+	}
+	if (match->mask.enc_ip_tos) {
+		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on IP ToS not supported");
+		return -EOPNOTSUPP;
+	}
+	if (match->mask.enc_ip_ttl) {
+		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on IP TTL not supported");
+		return -EOPNOTSUPP;
+	}
+
+	rc = efx_mae_check_encap_match_caps(efx, ipv6, extack);
+	if (rc) {
+		NL_SET_ERR_MSG_FMT_MOD(extack, "MAE hw reports no support for IPv%d encap matches",
+				       ipv6 ? 6 : 4);
+		return -EOPNOTSUPP;
+	}
+
+	encap = kzalloc(sizeof(*encap), GFP_USER);
+	if (!encap)
+		return -ENOMEM;
+	encap->src_ip = match->value.enc_src_ip;
+	encap->dst_ip = match->value.enc_dst_ip;
+#ifdef CONFIG_IPV6
+	encap->src_ip6 = match->value.enc_src_ip6;
+	encap->dst_ip6 = match->value.enc_dst_ip6;
+#endif
+	encap->udp_dport = match->value.enc_dport;
+	encap->tun_type = type;
+	old = rhashtable_lookup_get_insert_fast(&efx->tc->encap_match_ht,
+						&encap->linkage,
+						efx_tc_encap_match_ht_params);
+	if (old) {
+		/* don't need our new entry */
+		kfree(encap);
+		if (old->tun_type != type) {
+			NL_SET_ERR_MSG_FMT_MOD(extack,
+					       "Egress encap match with conflicting tun_type %u != %u",
+					       old->tun_type, type);
+			return -EEXIST;
+		}
+		if (!refcount_inc_not_zero(&old->ref))
+			return -EAGAIN;
+		/* existing entry found */
+		encap = old;
+	} else {
+		rc = efx_mae_register_encap_match(efx, encap);
+		if (rc) {
+			NL_SET_ERR_MSG_MOD(extack, "Failed to record egress encap match in HW");
+			goto fail;
+		}
+		refcount_set(&encap->ref, 1);
+	}
+	match->encap = encap;
+	return 0;
+fail:
+	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
+			       efx_tc_encap_match_ht_params);
+	kfree(encap);
+	return rc;
+}
+
+__always_unused
+static void efx_tc_flower_release_encap_match(struct efx_nic *efx,
+					      struct efx_tc_encap_match *encap)
+{
+	int rc;
+
+	if (!refcount_dec_and_test(&encap->ref))
+		return; /* still in use */
+
+	rc = efx_mae_unregister_encap_match(efx, encap);
+	if (rc)
+		/* Display message but carry on and remove entry from our
+		 * SW tables, because there's not much we can do about it.
+		 */
+		netif_err(efx, drv, efx->net_dev,
+			  "Failed to release encap match %#x, rc %d\n",
+			  encap->fw_id, rc);
+	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
+			       efx_tc_encap_match_ht_params);
+	kfree(encap);
+}
+
 /* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
 enum efx_tc_action_order {
 	EFX_TC_AO_VLAN_POP,

@@ -974,6 +1118,18 @@ void efx_fini_tc(struct efx_nic *efx)
 	efx->tc->up = false;
 }
 
+/* At teardown time, all TC filter rules (and thus all resources they created)
+ * should already have been removed.  If we find any in our hashtables, make a
+ * cursory attempt to clean up the software side.
+ */
+static void efx_tc_encap_match_free(void *ptr, void *__unused)
+{
+	struct efx_tc_encap_match *encap = ptr;
+
+	WARN_ON(refcount_read(&encap->ref));
+	kfree(encap);
+}
+
 int efx_init_struct_tc(struct efx_nic *efx)
 {
 	int rc;

@@ -996,6 +1152,9 @@ int efx_init_struct_tc(struct efx_nic *efx)
 	rc = efx_tc_init_counters(efx);
 	if (rc < 0)
 		goto fail_counters;
+	rc = rhashtable_init(&efx->tc->encap_match_ht, &efx_tc_encap_match_ht_params);
+	if (rc < 0)
+		goto fail_encap_match_ht;
 	rc = rhashtable_init(&efx->tc->match_action_ht, &efx_tc_match_action_ht_params);
 	if (rc < 0)
 		goto fail_match_action_ht;

@@ -1008,6 +1167,8 @@ int efx_init_struct_tc(struct efx_nic *efx)
 	efx->extra_channel_type[EFX_EXTRA_CHANNEL_TC] = &efx_tc_channel_type;
 	return 0;
 fail_match_action_ht:
+	rhashtable_destroy(&efx->tc->encap_match_ht);
+fail_encap_match_ht:
 	efx_tc_destroy_counters(efx);
 fail_counters:
 	mutex_destroy(&efx->tc->mutex);

@@ -1030,6 +1191,8 @@ void efx_fini_struct_tc(struct efx_nic *efx)
 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
 	rhashtable_free_and_destroy(&efx->tc->match_action_ht, efx_tc_flow_free,
 				    efx);
+	rhashtable_free_and_destroy(&efx->tc->encap_match_ht,
+				    efx_tc_encap_match_free, NULL);
 	efx_tc_fini_counters(efx);
 	mutex_unlock(&efx->tc->mutex);
 	mutex_destroy(&efx->tc->mutex);

index 19782c9a4354..d70c0ba86669 100644
@@ -18,6 +18,13 @@
 
 #define IS_ALL_ONES(v)	(!(typeof (v))~(v))
 
+#ifdef CONFIG_IPV6
+static inline bool efx_ipv6_addr_all_ones(struct in6_addr *addr)
+{
+	return !memchr_inv(addr, 0xff, sizeof(*addr));
+}
+#endif
+
 struct efx_tc_action_set {
 	u16 vlan_push:2;
 	u16 vlan_pop:2;

@@ -70,7 +77,9 @@ struct efx_tc_encap_match {
 	__be32 src_ip, dst_ip;
 	struct in6_addr src_ip6, dst_ip6;
 	__be16 udp_dport;
+	struct rhash_head linkage;
 	u16 tun_type; /* enum efx_encap_type */
+	refcount_t ref;
 	u32 fw_id; /* index of this entry in firmware encap match table */
 };
 

@@ -107,6 +116,7 @@ enum efx_tc_rule_prios {
  * @mutex: Used to serialise operations on TC hashtables
  * @counter_ht: Hashtable of TC counters (FW IDs and counter values)
  * @counter_id_ht: Hashtable mapping TC counter cookies to counters
+ * @encap_match_ht: Hashtable of TC encap matches
  * @match_action_ht: Hashtable of TC match-action rules
  * @reps_mport_id: MAE port allocated for representor RX
  * @reps_filter_uc: VNIC filter for representor unicast RX (promisc)

@@ -130,6 +140,7 @@ struct efx_tc_state {
 	struct mutex mutex;
 	struct rhashtable counter_ht;
 	struct rhashtable counter_id_ht;
+	struct rhashtable encap_match_ht;
 	struct rhashtable match_action_ht;
 	u32 reps_mport_id, reps_mport_vport_id;
 	s32 reps_filter_uc, reps_filter_mc;

index 92f1383ee4b9..49706a7b94bf 100644
@@ -241,6 +241,7 @@ static int efx_mae_get_basic_caps(struct efx_nic *efx, struct mae_caps *caps)
 	if (outlen < sizeof(outbuf))
 		return -EIO;
 	caps->match_field_count = MCDI_DWORD(outbuf, MAE_GET_CAPS_OUT_MATCH_FIELD_COUNT);
+	caps->encap_types = MCDI_DWORD(outbuf, MAE_GET_CAPS_OUT_ENCAP_TYPES_SUPPORTED);
 	caps->action_prios = MCDI_DWORD(outbuf, MAE_GET_CAPS_OUT_ACTION_PRIOS);
 	return 0;
 }

@@ -507,6 +508,25 @@ int efx_mae_check_encap_match_caps(struct efx_nic *efx, bool ipv6,
 }
 #undef CHECK
 
+int efx_mae_check_encap_type_supported(struct efx_nic *efx, enum efx_encap_type typ)
+{
+	unsigned int bit;
+
+	switch (typ & EFX_ENCAP_TYPES_MASK) {
+	case EFX_ENCAP_TYPE_VXLAN:
+		bit = MC_CMD_MAE_GET_CAPS_OUT_ENCAP_TYPE_VXLAN_LBN;
+		break;
+	case EFX_ENCAP_TYPE_GENEVE:
+		bit = MC_CMD_MAE_GET_CAPS_OUT_ENCAP_TYPE_GENEVE_LBN;
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+	if (efx->tc->caps->encap_types & BIT(bit))
+		return 0;
+	return -EOPNOTSUPP;
+}
+
 int efx_mae_allocate_counter(struct efx_nic *efx, struct efx_tc_counter *cnt)
 {
 	MCDI_DECLARE_BUF(outbuf, MC_CMD_MAE_COUNTER_ALLOC_OUT_LEN(1));

@@ -766,9 +786,10 @@ int efx_mae_alloc_action_set(struct efx_nic *efx, struct efx_tc_action_set *act)
 	size_t outlen;
 	int rc;
 
-	MCDI_POPULATE_DWORD_2(inbuf, MAE_ACTION_SET_ALLOC_IN_FLAGS,
+	MCDI_POPULATE_DWORD_3(inbuf, MAE_ACTION_SET_ALLOC_IN_FLAGS,
 			      MAE_ACTION_SET_ALLOC_IN_VLAN_PUSH, act->vlan_push,
-			      MAE_ACTION_SET_ALLOC_IN_VLAN_POP, act->vlan_pop);
+			      MAE_ACTION_SET_ALLOC_IN_VLAN_POP, act->vlan_pop,
+			      MAE_ACTION_SET_ALLOC_IN_DECAP, act->decap);
 
 	MCDI_SET_DWORD(inbuf, MAE_ACTION_SET_ALLOC_IN_SRC_MAC_ID,
 		       MC_CMD_MAE_MAC_ADDR_ALLOC_OUT_MAC_ID_NULL);

index 79d315a129ab..9226219491a0 100644
@@ -70,6 +70,7 @@ void efx_mae_counters_grant_credits(struct work_struct *work);
 
 struct mae_caps {
 	u32 match_field_count;
+	u32 encap_types;
 	u32 action_prios;
 	u8 action_rule_fields[MAE_NUM_FIELDS];
 	u8 outer_rule_fields[MAE_NUM_FIELDS];

@@ -82,6 +83,8 @@ int efx_mae_match_check_caps(struct efx_nic *efx,
 			     struct netlink_ext_ack *extack);
 int efx_mae_check_encap_match_caps(struct efx_nic *efx, bool ipv6,
 				   struct netlink_ext_ack *extack);
+int efx_mae_check_encap_type_supported(struct efx_nic *efx,
+				       enum efx_encap_type typ);
 
 int efx_mae_allocate_counter(struct efx_nic *efx, struct efx_tc_counter *cnt);
 int efx_mae_free_counter(struct efx_nic *efx, struct efx_tc_counter *cnt);

416 lines not on this page · raw message

index d70c0ba86669..47b6e9e35808 100644
@@ -28,6 +28,7 @@ static inline bool efx_ipv6_addr_all_ones(struct in6_addr *addr)
 struct efx_tc_action_set {
 	u16 vlan_push:2;
 	u16 vlan_pop:2;
+	u16 decap:1;
 	u16 deliver:1;
 	__be16 vlan_tci[2]; /* TCIs for vlan_push */
 	__be16 vlan_proto[2]; /* Ethertypes for vlan_push */