Merge branch 'sfc-tc-decap-support'

Edward Cree says:

====================
sfc: support TC decap rules

This series adds support for offloading tunnel decapsulation TC rules to
 ef100 NICs, allowing matching encapsulated packets to be decapsulated in
 hardware and redirected to VFs.
For now an encap match must be on precisely the following fields:
 ethertype (IPv4 or IPv6), source IP, destination IP, ipproto UDP,
 UDP destination port.  This simplifies checking for overlaps in the
 driver; the hardware supports a wider range of match fields which
 future driver work may expose.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/sfc/mae.c

@@ -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;
 }
@@ -254,13 +255,23 @@ 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 +289,12 @@ 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 +447,86 @@ 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_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));
@@ -488,6 +578,20 @@ 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;
@@ -682,9 +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);
@@ -845,6 +950,97 @@ 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)
 {
@@ -941,6 +1137,29 @@ 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;
 }
 

drivers/net/ethernet/sfc/mae.h

@@ -70,8 +70,10 @@ 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];
 };
 
 int efx_mae_get_caps(struct efx_nic *efx, struct mae_caps *caps);
@@ -79,6 +81,10 @@ 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_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);
@@ -91,6 +97,11 @@ 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);

drivers/net/ethernet/sfc/tc.c

@@ -10,12 +10,24 @@
  */
 
 #include <net/pkt_cls.h>
+#include <net/vxlan.h>
+#include <net/geneve.h>
 #include "tc.h"
 #include "tc_bindings.h"
 #include "mae.h"
 #include "ef100_rep.h"
 #include "efx.h"
 
+static enum efx_encap_type efx_tc_indr_netdev_type(struct net_device *net_dev)
+{
+	if (netif_is_vxlan(net_dev))
+		return EFX_ENCAP_TYPE_VXLAN;
+	if (netif_is_geneve(net_dev))
+		return EFX_ENCAP_TYPE_GENEVE;
+
+	return EFX_ENCAP_TYPE_NONE;
+}
+
 #define EFX_EFV_PF	NULL
 /* Look up the representor information (efv) for a device.
  * May return NULL for the PF (us), or an error pointer for a device that
@@ -43,6 +55,20 @@ static struct efx_rep *efx_tc_flower_lookup_efv(struct efx_nic *efx,
 	return efv;
 }
 
+/* Convert a driver-internal vport ID into an internal device (PF or VF) */
+static s64 efx_tc_flower_internal_mport(struct efx_nic *efx, struct efx_rep *efv)
+{
+	u32 mport;
+
+	if (IS_ERR(efv))
+		return PTR_ERR(efv);
+	if (!efv) /* device is PF (us) */
+		efx_mae_mport_uplink(efx, &mport);
+	else /* device is repr */
+		efx_mae_mport_mport(efx, efv->mport, &mport);
+	return mport;
+}
+
 /* Convert a driver-internal vport ID into an external device (wire or VF) */
 static s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv)
 {
@@ -57,6 +83,12 @@ 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),
@@ -66,7 +98,7 @@ 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.
 	 */
@@ -100,15 +132,6 @@ static void efx_tc_free_action_set_list(struct efx_nic *efx,
 	/* 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 void efx_tc_flow_free(void *ptr, void *arg)
 {
 	struct efx_tc_flow_rule *rule = ptr;
@@ -193,6 +216,11 @@ 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",
@@ -280,12 +308,226 @@ 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;
+}
+
+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 = false;
+	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;
+}
+
+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);
+}
+
+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);
+	if (rule->match.encap)
+		efx_tc_flower_release_encap_match(efx, rule->match.encap);
+	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
+}
+
+static const char *efx_tc_encap_type_name(enum efx_encap_type typ)
+{
+	switch (typ) {
+	case EFX_ENCAP_TYPE_NONE:
+		return "none";
+	case EFX_ENCAP_TYPE_VXLAN:
+		return "vxlan";
+	case EFX_ENCAP_TYPE_GENEVE:
+		return "geneve";
+	default:
+		pr_warn_once("Unknown efx_encap_type %d encountered\n", typ);
+		return "unknown";
+	}
 }
 
 /* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
 enum efx_tc_action_order {
+	EFX_TC_AO_DECAP,
 	EFX_TC_AO_VLAN_POP,
 	EFX_TC_AO_VLAN_PUSH,
 	EFX_TC_AO_COUNT,
@@ -296,6 +538,10 @@ static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
 					  enum efx_tc_action_order new)
 {
 	switch (new) {
+	case EFX_TC_AO_DECAP:
+		if (act->decap)
+			return false;
+		fallthrough;
 	case EFX_TC_AO_VLAN_POP:
 		if (act->vlan_pop >= 2)
 			return false;
@@ -323,6 +569,286 @@ static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
 	}
 }
 
+static int efx_tc_flower_replace_foreign(struct efx_nic *efx,
+					 struct net_device *net_dev,
+					 struct flow_cls_offload *tc)
+{
+	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
+	struct netlink_ext_ack *extack = tc->common.extack;
+	struct efx_tc_flow_rule *rule = NULL, *old = NULL;
+	struct efx_tc_action_set *act = NULL;
+	bool found = false, uplinked = false;
+	const struct flow_action_entry *fa;
+	struct efx_tc_match match;
+	struct efx_rep *to_efv;
+	s64 rc;
+	int i;
+
+	/* Parse match */
+	memset(&match, 0, sizeof(match));
+	rc = efx_tc_flower_parse_match(efx, fr, &match, NULL);
+	if (rc)
+		return rc;
+	/* The rule as given to us doesn't specify a source netdevice.
+	 * But, determining whether packets from a VF should match it is
+	 * complicated, so leave those to the software slowpath: qualify
+	 * the filter with source m-port == wire.
+	 */
+	rc = efx_tc_flower_external_mport(efx, EFX_EFV_PF);
+	if (rc < 0) {
+		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port for foreign filter");
+		return rc;
+	}
+	match.value.ingress_port = rc;
+	match.mask.ingress_port = ~0;
+
+	if (tc->common.chain_index) {
+		NL_SET_ERR_MSG_MOD(extack, "No support for nonzero chain_index");
+		return -EOPNOTSUPP;
+	}
+	match.mask.recirc_id = 0xff;
+
+	flow_action_for_each(i, fa, &fr->action) {
+		switch (fa->id) {
+		case FLOW_ACTION_REDIRECT:
+		case FLOW_ACTION_MIRRED: /* mirred means mirror here */
+			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
+			if (IS_ERR(to_efv))
+				continue;
+			found = true;
+			break;
+		default:
+			break;
+		}
+	}
+	if (!found) { /* We don't care. */
+		netif_dbg(efx, drv, efx->net_dev,
+			  "Ignoring foreign filter that doesn't egdev us\n");
+		rc = -EOPNOTSUPP;
+		goto release;
+	}
+
+	rc = efx_mae_match_check_caps(efx, &match.mask, NULL);
+	if (rc)
+		goto release;
+
+	if (efx_tc_match_is_encap(&match.mask)) {
+		enum efx_encap_type type;
+
+		type = efx_tc_indr_netdev_type(net_dev);
+		if (type == EFX_ENCAP_TYPE_NONE) {
+			NL_SET_ERR_MSG_MOD(extack,
+					   "Egress encap match on unsupported tunnel device");
+			rc = -EOPNOTSUPP;
+			goto release;
+		}
+
+		rc = efx_mae_check_encap_type_supported(efx, type);
+		if (rc) {
+			NL_SET_ERR_MSG_FMT_MOD(extack,
+					       "Firmware reports no support for %s encap match",
+					       efx_tc_encap_type_name(type));
+			goto release;
+		}
+
+		rc = efx_tc_flower_record_encap_match(efx, &match, type,
+						      extack);
+		if (rc)
+			goto release;
+	} else {
+		/* This is not a tunnel decap rule, ignore it */
+		netif_dbg(efx, drv, efx->net_dev,
+			  "Ignoring foreign filter without encap match\n");
+		rc = -EOPNOTSUPP;
+		goto release;
+	}
+
+	rule = kzalloc(sizeof(*rule), GFP_USER);
+	if (!rule) {
+		rc = -ENOMEM;
+		goto release;
+	}
+	INIT_LIST_HEAD(&rule->acts.list);
+	rule->cookie = tc->cookie;
+	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
+						&rule->linkage,
+						efx_tc_match_action_ht_params);
+	if (old) {
+		netif_dbg(efx, drv, efx->net_dev,
+			  "Ignoring already-offloaded rule (cookie %lx)\n",
+			  tc->cookie);
+		rc = -EEXIST;
+		goto release;
+	}
+
+	act = kzalloc(sizeof(*act), GFP_USER);
+	if (!act) {
+		rc = -ENOMEM;
+		goto release;
+	}
+
+	/* Parse actions.  For foreign rules we only support decap & redirect.
+	 * See corresponding code in efx_tc_flower_replace() for theory of
+	 * operation & how 'act' cursor is used.
+	 */
+	flow_action_for_each(i, fa, &fr->action) {
+		struct efx_tc_action_set save;
+
+		switch (fa->id) {
+		case FLOW_ACTION_REDIRECT:
+		case FLOW_ACTION_MIRRED:
+			/* See corresponding code in efx_tc_flower_replace() for
+			 * long explanations of what's going on here.
+			 */
+			save = *act;
+			if (fa->hw_stats) {
+				struct efx_tc_counter_index *ctr;
+
+				if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
+					NL_SET_ERR_MSG_FMT_MOD(extack,
+							       "hw_stats_type %u not supported (only 'delayed')",
+							       fa->hw_stats);
+					rc = -EOPNOTSUPP;
+					goto release;
+				}
+				if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
+					rc = -EOPNOTSUPP;
+					goto release;
+				}
+
+				ctr = efx_tc_flower_get_counter_index(efx,
+								      tc->cookie,
+								      EFX_TC_COUNTER_TYPE_AR);
+				if (IS_ERR(ctr)) {
+					rc = PTR_ERR(ctr);
+					NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
+					goto release;
+				}
+				act->count = ctr;
+			}
+
+			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
+				/* can't happen */
+				rc = -EOPNOTSUPP;
+				NL_SET_ERR_MSG_MOD(extack,
+						   "Deliver action violates action order (can't happen)");
+				goto release;
+			}
+			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
+			/* PF implies egdev is us, in which case we really
+			 * want to deliver to the uplink (because this is an
+			 * ingress filter).  If we don't recognise the egdev
+			 * at all, then we'd better trap so SW can handle it.
+			 */
+			if (IS_ERR(to_efv))
+				to_efv = EFX_EFV_PF;
+			if (to_efv == EFX_EFV_PF) {
+				if (uplinked)
+					break;
+				uplinked = true;
+			}
+			rc = efx_tc_flower_internal_mport(efx, to_efv);
+			if (rc < 0) {
+				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
+				goto release;
+			}
+			act->dest_mport = rc;
+			act->deliver = 1;
+			rc = efx_mae_alloc_action_set(efx, act);
+			if (rc) {
+				NL_SET_ERR_MSG_MOD(extack,
+						   "Failed to write action set to hw (mirred)");
+				goto release;
+			}
+			list_add_tail(&act->list, &rule->acts.list);
+			act = NULL;
+			if (fa->id == FLOW_ACTION_REDIRECT)
+				break; /* end of the line */
+			/* Mirror, so continue on with saved act */
+			act = kzalloc(sizeof(*act), GFP_USER);
+			if (!act) {
+				rc = -ENOMEM;
+				goto release;
+			}
+			*act = save;
+			break;
+		case FLOW_ACTION_TUNNEL_DECAP:
+			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DECAP)) {
+				rc = -EINVAL;
+				NL_SET_ERR_MSG_MOD(extack, "Decap action violates action order");
+				goto release;
+			}
+			act->decap = 1;
+			/* If we previously delivered/trapped to uplink, now
+			 * that we've decapped we'll want another copy if we
+			 * try to deliver/trap to uplink again.
+			 */
+			uplinked = false;
+			break;
+		default:
+			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
+					       fa->id);
+			rc = -EOPNOTSUPP;
+			goto release;
+		}
+	}
+
+	if (act) {
+		if (!uplinked) {
+			/* Not shot/redirected, so deliver to default dest (which is
+			 * the uplink, as this is an ingress filter)
+			 */
+			efx_mae_mport_uplink(efx, &act->dest_mport);
+			act->deliver = 1;
+		}
+		rc = efx_mae_alloc_action_set(efx, act);
+		if (rc) {
+			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
+			goto release;
+		}
+		list_add_tail(&act->list, &rule->acts.list);
+		act = NULL; /* Prevent double-free in error path */
+	}
+
+	rule->match = match;
+
+	netif_dbg(efx, drv, efx->net_dev,
+		  "Successfully parsed foreign filter (cookie %lx)\n",
+		  tc->cookie);
+
+	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
+	if (rc) {
+		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
+		goto release;
+	}
+	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
+				 rule->acts.fw_id, &rule->fw_id);
+	if (rc) {
+		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
+		goto release_acts;
+	}
+	return 0;
+
+release_acts:
+	efx_mae_free_action_set_list(efx, &rule->acts);
+release:
+	/* We failed to insert the rule, so free up any entries we created in
+	 * subsidiary tables.
+	 */
+	if (act)
+		efx_tc_free_action_set(efx, act, false);
+	if (rule) {
+		rhashtable_remove_fast(&efx->tc->match_action_ht,
+				       &rule->linkage,
+				       efx_tc_match_action_ht_params);
+		efx_tc_free_action_set_list(efx, &rule->acts, false);
+	}
+	kfree(rule);
+	if (match.encap)
+		efx_tc_flower_release_encap_match(efx, match.encap);
+	return rc;
+}
+
 static int efx_tc_flower_replace(struct efx_nic *efx,
 				 struct net_device *net_dev,
 				 struct flow_cls_offload *tc,
@@ -347,10 +873,8 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
 
 	from_efv = efx_tc_flower_lookup_efv(efx, net_dev);
 	if (IS_ERR(from_efv)) {
-		/* Might be a tunnel decap rule from an indirect block.
-		 * Support for those not implemented yet.
-		 */
-		return -EOPNOTSUPP;
+		/* Not from our PF or representors, so probably a tunnel dev */
+		return efx_tc_flower_replace_foreign(efx, net_dev, tc);
 	}
 
 	if (efv != from_efv) {
@@ -373,6 +897,11 @@ 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");
@@ -407,6 +936,30 @@ 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;
@@ -889,6 +1442,18 @@ 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;
@@ -911,6 +1476,9 @@ 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;
@@ -923,6 +1491,8 @@ 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);
@@ -945,6 +1515,8 @@ 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);

drivers/net/ethernet/sfc/tc.h

@@ -18,9 +18,17 @@
 
 #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;
+	u16 decap:1;
 	u16 deliver:1;
 	__be16 vlan_tci[2]; /* TCIs for vlan_push */
 	__be16 vlan_proto[2]; /* Ethertypes for vlan_push */
@@ -48,11 +56,38 @@ 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;
+	struct rhash_head linkage;
+	enum efx_encap_type tun_type;
+	refcount_t ref;
+	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 {
@@ -82,6 +117,7 @@ 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)
@@ -105,6 +141,7 @@ 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;