1 files changed, 609 insertions, 9 deletions

diff --git a/drivers/net/ethernet/netronome/nfp/flower/offload.c b/drivers/net/ethernet/netronome/nfp/flower/offload.c
index 450d7296fd57..1fbfeb43c538 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/offload.c
@@ -55,6 +55,28 @@
 	 BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) | \
 	 BIT(FLOW_DISSECTOR_KEY_ENC_PORTS))
 
+#define NFP_FLOWER_MERGE_FIELDS \
+	(NFP_FLOWER_LAYER_PORT | \
+	 NFP_FLOWER_LAYER_MAC | \
+	 NFP_FLOWER_LAYER_TP | \
+	 NFP_FLOWER_LAYER_IPV4 | \
+	 NFP_FLOWER_LAYER_IPV6)
+
+struct nfp_flower_merge_check {
+	union {
+		struct {
+			__be16 tci;
+			struct nfp_flower_mac_mpls l2;
+			struct nfp_flower_tp_ports l4;
+			union {
+				struct nfp_flower_ipv4 ipv4;
+				struct nfp_flower_ipv6 ipv6;
+			};
+		};
+		unsigned long vals[8];
+	};
+};
+
 static int
 nfp_flower_xmit_flow(struct nfp_app *app, struct nfp_fl_payload *nfp_flow,
 		     u8 mtype)
@@ -195,7 +217,7 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 			flow_rule_match_enc_opts(rule, &enc_op);
 
 		switch (enc_ports.key->dst) {
-		case htons(NFP_FL_VXLAN_PORT):
+		case htons(IANA_VXLAN_UDP_PORT):
 			*tun_type = NFP_FL_TUNNEL_VXLAN;
 			key_layer |= NFP_FLOWER_LAYER_VXLAN;
 			key_size += sizeof(struct nfp_flower_ipv4_udp_tun);
@@ -203,7 +225,7 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 			if (enc_op.key)
 				return -EOPNOTSUPP;
 			break;
-		case htons(NFP_FL_GENEVE_PORT):
+		case htons(GENEVE_UDP_PORT):
 			if (!(priv->flower_ext_feats & NFP_FL_FEATS_GENEVE))
 				return -EOPNOTSUPP;
 			*tun_type = NFP_FL_TUNNEL_GENEVE;
@@ -326,7 +348,7 @@ nfp_flower_calculate_key_layers(struct nfp_app *app,
 				break;
 
 			case cpu_to_be16(ETH_P_IPV6):
-				key_layer |= NFP_FLOWER_LAYER_IPV6;
+					key_layer |= NFP_FLOWER_LAYER_IPV6;
 				key_size += sizeof(struct nfp_flower_ipv6);
 				break;
 
@@ -376,6 +398,8 @@ nfp_flower_allocate_new(struct nfp_fl_key_ls *key_layer)
 
 	flow_pay->nfp_tun_ipv4_addr = 0;
 	flow_pay->meta.flags = 0;
+	INIT_LIST_HEAD(&flow_pay->linked_flows);
+	flow_pay->in_hw = false;
 
 	return flow_pay;
 
@@ -388,6 +412,447 @@ err_free_flow:
 	return NULL;
 }
 
+static int
+nfp_flower_update_merge_with_actions(struct nfp_fl_payload *flow,
+				     struct nfp_flower_merge_check *merge,
+				     u8 *last_act_id, int *act_out)
+{
+	struct nfp_fl_set_ipv6_tc_hl_fl *ipv6_tc_hl_fl;
+	struct nfp_fl_set_ip4_ttl_tos *ipv4_ttl_tos;
+	struct nfp_fl_set_ip4_addrs *ipv4_add;
+	struct nfp_fl_set_ipv6_addr *ipv6_add;
+	struct nfp_fl_push_vlan *push_vlan;
+	struct nfp_fl_set_tport *tport;
+	struct nfp_fl_set_eth *eth;
+	struct nfp_fl_act_head *a;
+	unsigned int act_off = 0;
+	u8 act_id = 0;
+	u8 *ports;
+	int i;
+
+	while (act_off < flow->meta.act_len) {
+		a = (struct nfp_fl_act_head *)&flow->action_data[act_off];
+		act_id = a->jump_id;
+
+		switch (act_id) {
+		case NFP_FL_ACTION_OPCODE_OUTPUT:
+			if (act_out)
+				(*act_out)++;
+			break;
+		case NFP_FL_ACTION_OPCODE_PUSH_VLAN:
+			push_vlan = (struct nfp_fl_push_vlan *)a;
+			if (push_vlan->vlan_tci)
+				merge->tci = cpu_to_be16(0xffff);
+			break;
+		case NFP_FL_ACTION_OPCODE_POP_VLAN:
+			merge->tci = cpu_to_be16(0);
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV4_TUNNEL:
+			/* New tunnel header means l2 to l4 can be matched. */
+			eth_broadcast_addr(&merge->l2.mac_dst[0]);
+			eth_broadcast_addr(&merge->l2.mac_src[0]);
+			memset(&merge->l4, 0xff,
+			       sizeof(struct nfp_flower_tp_ports));
+			memset(&merge->ipv4, 0xff,
+			       sizeof(struct nfp_flower_ipv4));
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_ETHERNET:
+			eth = (struct nfp_fl_set_eth *)a;
+			for (i = 0; i < ETH_ALEN; i++)
+				merge->l2.mac_dst[i] |= eth->eth_addr_mask[i];
+			for (i = 0; i < ETH_ALEN; i++)
+				merge->l2.mac_src[i] |=
+					eth->eth_addr_mask[ETH_ALEN + i];
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV4_ADDRS:
+			ipv4_add = (struct nfp_fl_set_ip4_addrs *)a;
+			merge->ipv4.ipv4_src |= ipv4_add->ipv4_src_mask;
+			merge->ipv4.ipv4_dst |= ipv4_add->ipv4_dst_mask;
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV4_TTL_TOS:
+			ipv4_ttl_tos = (struct nfp_fl_set_ip4_ttl_tos *)a;
+			merge->ipv4.ip_ext.ttl |= ipv4_ttl_tos->ipv4_ttl_mask;
+			merge->ipv4.ip_ext.tos |= ipv4_ttl_tos->ipv4_tos_mask;
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV6_SRC:
+			ipv6_add = (struct nfp_fl_set_ipv6_addr *)a;
+			for (i = 0; i < 4; i++)
+				merge->ipv6.ipv6_src.in6_u.u6_addr32[i] |=
+					ipv6_add->ipv6[i].mask;
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV6_DST:
+			ipv6_add = (struct nfp_fl_set_ipv6_addr *)a;
+			for (i = 0; i < 4; i++)
+				merge->ipv6.ipv6_dst.in6_u.u6_addr32[i] |=
+					ipv6_add->ipv6[i].mask;
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_IPV6_TC_HL_FL:
+			ipv6_tc_hl_fl = (struct nfp_fl_set_ipv6_tc_hl_fl *)a;
+			merge->ipv6.ip_ext.ttl |=
+				ipv6_tc_hl_fl->ipv6_hop_limit_mask;
+			merge->ipv6.ip_ext.tos |= ipv6_tc_hl_fl->ipv6_tc_mask;
+			merge->ipv6.ipv6_flow_label_exthdr |=
+				ipv6_tc_hl_fl->ipv6_label_mask;
+			break;
+		case NFP_FL_ACTION_OPCODE_SET_UDP:
+		case NFP_FL_ACTION_OPCODE_SET_TCP:
+			tport = (struct nfp_fl_set_tport *)a;
+			ports = (u8 *)&merge->l4.port_src;
+			for (i = 0; i < 4; i++)
+				ports[i] |= tport->tp_port_mask[i];
+			break;
+		case NFP_FL_ACTION_OPCODE_PRE_TUNNEL:
+		case NFP_FL_ACTION_OPCODE_PRE_LAG:
+		case NFP_FL_ACTION_OPCODE_PUSH_GENEVE:
+			break;
+		default:
+			return -EOPNOTSUPP;
+		}
+
+		act_off += a->len_lw << NFP_FL_LW_SIZ;
+	}
+
+	if (last_act_id)
+		*last_act_id = act_id;
+
+	return 0;
+}
+
+static int
+nfp_flower_populate_merge_match(struct nfp_fl_payload *flow,
+				struct nfp_flower_merge_check *merge,
+				bool extra_fields)
+{
+	struct nfp_flower_meta_tci *meta_tci;
+	u8 *mask = flow->mask_data;
+	u8 key_layer, match_size;
+
+	memset(merge, 0, sizeof(struct nfp_flower_merge_check));
+
+	meta_tci = (struct nfp_flower_meta_tci *)mask;
+	key_layer = meta_tci->nfp_flow_key_layer;
+
+	if (key_layer & ~NFP_FLOWER_MERGE_FIELDS && !extra_fields)
+		return -EOPNOTSUPP;
+
+	merge->tci = meta_tci->tci;
+	mask += sizeof(struct nfp_flower_meta_tci);
+
+	if (key_layer & NFP_FLOWER_LAYER_EXT_META)
+		mask += sizeof(struct nfp_flower_ext_meta);
+
+	mask += sizeof(struct nfp_flower_in_port);
+
+	if (key_layer & NFP_FLOWER_LAYER_MAC) {
+		match_size = sizeof(struct nfp_flower_mac_mpls);
+		memcpy(&merge->l2, mask, match_size);
+		mask += match_size;
+	}
+
+	if (key_layer & NFP_FLOWER_LAYER_TP) {
+		match_size = sizeof(struct nfp_flower_tp_ports);
+		memcpy(&merge->l4, mask, match_size);
+		mask += match_size;
+	}
+
+	if (key_layer & NFP_FLOWER_LAYER_IPV4) {
+		match_size = sizeof(struct nfp_flower_ipv4);
+		memcpy(&merge->ipv4, mask, match_size);
+	}
+
+	if (key_layer & NFP_FLOWER_LAYER_IPV6) {
+		match_size = sizeof(struct nfp_flower_ipv6);
+		memcpy(&merge->ipv6, mask, match_size);
+	}
+
+	return 0;
+}
+
+static int
+nfp_flower_can_merge(struct nfp_fl_payload *sub_flow1,
+		     struct nfp_fl_payload *sub_flow2)
+{
+	/* Two flows can be merged if sub_flow2 only matches on bits that are
+	 * either matched by sub_flow1 or set by a sub_flow1 action. This
+	 * ensures that every packet that hits sub_flow1 and recirculates is
+	 * guaranteed to hit sub_flow2.
+	 */
+	struct nfp_flower_merge_check sub_flow1_merge, sub_flow2_merge;
+	int err, act_out = 0;
+	u8 last_act_id = 0;
+
+	err = nfp_flower_populate_merge_match(sub_flow1, &sub_flow1_merge,
+					      true);
+	if (err)
+		return err;
+
+	err = nfp_flower_populate_merge_match(sub_flow2, &sub_flow2_merge,
+					      false);
+	if (err)
+		return err;
+
+	err = nfp_flower_update_merge_with_actions(sub_flow1, &sub_flow1_merge,
+						   &last_act_id, &act_out);
+	if (err)
+		return err;
+
+	/* Must only be 1 output action and it must be the last in sequence. */
+	if (act_out != 1 || last_act_id != NFP_FL_ACTION_OPCODE_OUTPUT)
+		return -EOPNOTSUPP;
+
+	/* Reject merge if sub_flow2 matches on something that is not matched
+	 * on or set in an action by sub_flow1.
+	 */
+	err = bitmap_andnot(sub_flow2_merge.vals, sub_flow2_merge.vals,
+			    sub_flow1_merge.vals,
+			    sizeof(struct nfp_flower_merge_check) * 8);
+	if (err)
+		return -EINVAL;
+
+	return 0;
+}
+
+static unsigned int
+nfp_flower_copy_pre_actions(char *act_dst, char *act_src, int len,
+			    bool *tunnel_act)
+{
+	unsigned int act_off = 0, act_len;
+	struct nfp_fl_act_head *a;
+	u8 act_id = 0;
+
+	while (act_off < len) {
+		a = (struct nfp_fl_act_head *)&act_src[act_off];
+		act_len = a->len_lw << NFP_FL_LW_SIZ;
+		act_id = a->jump_id;
+
+		switch (act_id) {
+		case NFP_FL_ACTION_OPCODE_PRE_TUNNEL:
+			if (tunnel_act)
+				*tunnel_act = true;
+			/* fall through */
+		case NFP_FL_ACTION_OPCODE_PRE_LAG:
+			memcpy(act_dst + act_off, act_src + act_off, act_len);
+			break;
+		default:
+			return act_off;
+		}
+
+		act_off += act_len;
+	}
+
+	return act_off;
+}
+
+static int nfp_fl_verify_post_tun_acts(char *acts, int len)
+{
+	struct nfp_fl_act_head *a;
+	unsigned int act_off = 0;
+
+	while (act_off < len) {
+		a = (struct nfp_fl_act_head *)&acts[act_off];
+		if (a->jump_id != NFP_FL_ACTION_OPCODE_OUTPUT)
+			return -EOPNOTSUPP;
+
+		act_off += a->len_lw << NFP_FL_LW_SIZ;
+	}
+
+	return 0;
+}
+
+static int
+nfp_flower_merge_action(struct nfp_fl_payload *sub_flow1,
+			struct nfp_fl_payload *sub_flow2,
+			struct nfp_fl_payload *merge_flow)
+{
+	unsigned int sub1_act_len, sub2_act_len, pre_off1, pre_off2;
+	bool tunnel_act = false;
+	char *merge_act;
+	int err;
+
+	/* The last action of sub_flow1 must be output - do not merge this. */
+	sub1_act_len = sub_flow1->meta.act_len - sizeof(struct nfp_fl_output);
+	sub2_act_len = sub_flow2->meta.act_len;
+
+	if (!sub2_act_len)
+		return -EINVAL;
+
+	if (sub1_act_len + sub2_act_len > NFP_FL_MAX_A_SIZ)
+		return -EINVAL;
+
+	/* A shortcut can only be applied if there is a single action. */
+	if (sub1_act_len)
+		merge_flow->meta.shortcut = cpu_to_be32(NFP_FL_SC_ACT_NULL);
+	else
+		merge_flow->meta.shortcut = sub_flow2->meta.shortcut;
+
+	merge_flow->meta.act_len = sub1_act_len + sub2_act_len;
+	merge_act = merge_flow->action_data;
+
+	/* Copy any pre-actions to the start of merge flow action list. */
+	pre_off1 = nfp_flower_copy_pre_actions(merge_act,
+					       sub_flow1->action_data,
+					       sub1_act_len, &tunnel_act);
+	merge_act += pre_off1;
+	sub1_act_len -= pre_off1;
+	pre_off2 = nfp_flower_copy_pre_actions(merge_act,
+					       sub_flow2->action_data,
+					       sub2_act_len, NULL);
+	merge_act += pre_off2;
+	sub2_act_len -= pre_off2;
+
+	/* FW does a tunnel push when egressing, therefore, if sub_flow 1 pushes
+	 * a tunnel, sub_flow 2 can only have output actions for a valid merge.
+	 */
+	if (tunnel_act) {
+		char *post_tun_acts = &sub_flow2->action_data[pre_off2];
+
+		err = nfp_fl_verify_post_tun_acts(post_tun_acts, sub2_act_len);
+		if (err)
+			return err;
+	}
+
+	/* Copy remaining actions from sub_flows 1 and 2. */
+	memcpy(merge_act, sub_flow1->action_data + pre_off1, sub1_act_len);
+	merge_act += sub1_act_len;
+	memcpy(merge_act, sub_flow2->action_data + pre_off2, sub2_act_len);
+
+	return 0;
+}
+
+/* Flow link code should only be accessed under RTNL. */
+static void nfp_flower_unlink_flow(struct nfp_fl_payload_link *link)
+{
+	list_del(&link->merge_flow.list);
+	list_del(&link->sub_flow.list);
+	kfree(link);
+}
+
+static void nfp_flower_unlink_flows(struct nfp_fl_payload *merge_flow,
+				    struct nfp_fl_payload *sub_flow)
+{
+	struct nfp_fl_payload_link *link;
+
+	list_for_each_entry(link, &merge_flow->linked_flows, merge_flow.list)
+		if (link->sub_flow.flow == sub_flow) {
+			nfp_flower_unlink_flow(link);
+			return;
+		}
+}
+
+static int nfp_flower_link_flows(struct nfp_fl_payload *merge_flow,
+				 struct nfp_fl_payload *sub_flow)
+{
+	struct nfp_fl_payload_link *link;
+
+	link = kmalloc(sizeof(*link), GFP_KERNEL);
+	if (!link)
+		return -ENOMEM;
+
+	link->merge_flow.flow = merge_flow;
+	list_add_tail(&link->merge_flow.list, &merge_flow->linked_flows);
+	link->sub_flow.flow = sub_flow;
+	list_add_tail(&link->sub_flow.list, &sub_flow->linked_flows);
+
+	return 0;
+}
+
+/**
+ * nfp_flower_merge_offloaded_flows() - Merge 2 existing flows to single flow.
+ * @app:	Pointer to the APP handle
+ * @sub_flow1:	Initial flow matched to produce merge hint
+ * @sub_flow2:	Post recirculation flow matched in merge hint
+ *
+ * Combines 2 flows (if valid) to a single flow, removing the initial from hw
+ * and offloading the new, merged flow.
+ *
+ * Return: negative value on error, 0 in success.
+ */
+int nfp_flower_merge_offloaded_flows(struct nfp_app *app,
+				     struct nfp_fl_payload *sub_flow1,
+				     struct nfp_fl_payload *sub_flow2)
+{
+	struct tc_cls_flower_offload merge_tc_off;
+	struct nfp_flower_priv *priv = app->priv;
+	struct nfp_fl_payload *merge_flow;
+	struct nfp_fl_key_ls merge_key_ls;
+	int err;
+
+	ASSERT_RTNL();
+
+	if (sub_flow1 == sub_flow2 ||
+	    nfp_flower_is_merge_flow(sub_flow1) ||
+	    nfp_flower_is_merge_flow(sub_flow2))
+		return -EINVAL;
+
+	err = nfp_flower_can_merge(sub_flow1, sub_flow2);
+	if (err)
+		return err;
+
+	merge_key_ls.key_size = sub_flow1->meta.key_len;
+
+	merge_flow = nfp_flower_allocate_new(&merge_key_ls);
+	if (!merge_flow)
+		return -ENOMEM;
+
+	merge_flow->tc_flower_cookie = (unsigned long)merge_flow;
+	merge_flow->ingress_dev = sub_flow1->ingress_dev;
+
+	memcpy(merge_flow->unmasked_data, sub_flow1->unmasked_data,
+	       sub_flow1->meta.key_len);
+	memcpy(merge_flow->mask_data, sub_flow1->mask_data,
+	       sub_flow1->meta.mask_len);
+
+	err = nfp_flower_merge_action(sub_flow1, sub_flow2, merge_flow);
+	if (err)
+		goto err_destroy_merge_flow;
+
+	err = nfp_flower_link_flows(merge_flow, sub_flow1);
+	if (err)
+		goto err_destroy_merge_flow;
+
+	err = nfp_flower_link_flows(merge_flow, sub_flow2);
+	if (err)
+		goto err_unlink_sub_flow1;
+
+	merge_tc_off.cookie = merge_flow->tc_flower_cookie;
+	err = nfp_compile_flow_metadata(app, &merge_tc_off, merge_flow,
+					merge_flow->ingress_dev);
+	if (err)
+		goto err_unlink_sub_flow2;
+
+	err = rhashtable_insert_fast(&priv->flow_table, &merge_flow->fl_node,
+				     nfp_flower_table_params);
+	if (err)
+		goto err_release_metadata;
+
+	err = nfp_flower_xmit_flow(app, merge_flow,
+				   NFP_FLOWER_CMSG_TYPE_FLOW_MOD);
+	if (err)
+		goto err_remove_rhash;
+
+	merge_flow->in_hw = true;
+	sub_flow1->in_hw = false;
+
+	return 0;
+
+err_remove_rhash:
+	WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
+					    &merge_flow->fl_node,
+					    nfp_flower_table_params));
+err_release_metadata:
+	nfp_modify_flow_metadata(app, merge_flow);
+err_unlink_sub_flow2:
+	nfp_flower_unlink_flows(merge_flow, sub_flow2);
+err_unlink_sub_flow1:
+	nfp_flower_unlink_flows(merge_flow, sub_flow1);
+err_destroy_merge_flow:
+	kfree(merge_flow->action_data);
+	kfree(merge_flow->mask_data);
+	kfree(merge_flow->unmasked_data);
+	kfree(merge_flow);
+	return err;
+}
+
 /**
  * nfp_flower_add_offload() - Adds a new flow to hardware.
  * @app:	Pointer to the APP handle
@@ -454,6 +919,8 @@ nfp_flower_add_offload(struct nfp_app *app, struct net_device *netdev,
 	if (port)
 		port->tc_offload_cnt++;
 
+	flow_pay->in_hw = true;
+
 	/* Deallocate flow payload when flower rule has been destroyed. */
 	kfree(key_layer);
 
@@ -475,6 +942,75 @@ err_free_key_ls:
 	return err;
 }
 
+static void
+nfp_flower_remove_merge_flow(struct nfp_app *app,
+			     struct nfp_fl_payload *del_sub_flow,
+			     struct nfp_fl_payload *merge_flow)
+{
+	struct nfp_flower_priv *priv = app->priv;
+	struct nfp_fl_payload_link *link, *temp;
+	struct nfp_fl_payload *origin;
+	bool mod = false;
+	int err;
+
+	link = list_first_entry(&merge_flow->linked_flows,
+				struct nfp_fl_payload_link, merge_flow.list);
+	origin = link->sub_flow.flow;
+
+	/* Re-add rule the merge had overwritten if it has not been deleted. */
+	if (origin != del_sub_flow)
+		mod = true;
+
+	err = nfp_modify_flow_metadata(app, merge_flow);
+	if (err) {
+		nfp_flower_cmsg_warn(app, "Metadata fail for merge flow delete.\n");
+		goto err_free_links;
+	}
+
+	if (!mod) {
+		err = nfp_flower_xmit_flow(app, merge_flow,
+					   NFP_FLOWER_CMSG_TYPE_FLOW_DEL);
+		if (err) {
+			nfp_flower_cmsg_warn(app, "Failed to delete merged flow.\n");
+			goto err_free_links;
+		}
+	} else {
+		__nfp_modify_flow_metadata(priv, origin);
+		err = nfp_flower_xmit_flow(app, origin,
+					   NFP_FLOWER_CMSG_TYPE_FLOW_MOD);
+		if (err)
+			nfp_flower_cmsg_warn(app, "Failed to revert merge flow.\n");
+		origin->in_hw = true;
+	}
+
+err_free_links:
+	/* Clean any links connected with the merged flow. */
+	list_for_each_entry_safe(link, temp, &merge_flow->linked_flows,
+				 merge_flow.list)
+		nfp_flower_unlink_flow(link);
+
+	kfree(merge_flow->action_data);
+	kfree(merge_flow->mask_data);
+	kfree(merge_flow->unmasked_data);
+	WARN_ON_ONCE(rhashtable_remove_fast(&priv->flow_table,
+					    &merge_flow->fl_node,
+					    nfp_flower_table_params));
+	kfree_rcu(merge_flow, rcu);
+}
+
+static void
+nfp_flower_del_linked_merge_flows(struct nfp_app *app,
+				  struct nfp_fl_payload *sub_flow)
+{
+	struct nfp_fl_payload_link *link, *temp;
+
+	/* Remove any merge flow formed from the deleted sub_flow. */
+	list_for_each_entry_safe(link, temp, &sub_flow->linked_flows,
+				 sub_flow.list)
+		nfp_flower_remove_merge_flow(app, sub_flow,
+					     link->merge_flow.flow);
+}
+
 /**
  * nfp_flower_del_offload() - Removes a flow from hardware.
  * @app:	Pointer to the APP handle
@@ -482,7 +1018,7 @@ err_free_key_ls:
  * @flow:	TC flower classifier offload structure
  *
  * Removes a flow from the repeated hash structure and clears the
- * action payload.
+ * action payload. Any flows merged from this are also deleted.
  *
  * Return: negative value on error, 0 if removed successfully.
  */
@@ -504,17 +1040,22 @@ nfp_flower_del_offload(struct nfp_app *app, struct net_device *netdev,
 
 	err = nfp_modify_flow_metadata(app, nfp_flow);
 	if (err)
-		goto err_free_flow;
+		goto err_free_merge_flow;
 
 	if (nfp_flow->nfp_tun_ipv4_addr)
 		nfp_tunnel_del_ipv4_off(app, nfp_flow->nfp_tun_ipv4_addr);
 
+	if (!nfp_flow->in_hw) {
+		err = 0;
+		goto err_free_merge_flow;
+	}
+
 	err = nfp_flower_xmit_flow(app, nfp_flow,
 				   NFP_FLOWER_CMSG_TYPE_FLOW_DEL);
-	if (err)
-		goto err_free_flow;
+	/* Fall through on error. */
 
-err_free_flow:
+err_free_merge_flow:
+	nfp_flower_del_linked_merge_flows(app, nfp_flow);
 	if (port)
 		port->tc_offload_cnt--;
 	kfree(nfp_flow->action_data);
@@ -527,6 +1068,52 @@ err_free_flow:
 	return err;
 }
 
+static void
+__nfp_flower_update_merge_stats(struct nfp_app *app,
+				struct nfp_fl_payload *merge_flow)
+{
+	struct nfp_flower_priv *priv = app->priv;
+	struct nfp_fl_payload_link *link;
+	struct nfp_fl_payload *sub_flow;
+	u64 pkts, bytes, used;
+	u32 ctx_id;
+
+	ctx_id = be32_to_cpu(merge_flow->meta.host_ctx_id);
+	pkts = priv->stats[ctx_id].pkts;
+	/* Do not cycle subflows if no stats to distribute. */
+	if (!pkts)
+		return;
+	bytes = priv->stats[ctx_id].bytes;
+	used = priv->stats[ctx_id].used;
+
+	/* Reset stats for the merge flow. */
+	priv->stats[ctx_id].pkts = 0;
+	priv->stats[ctx_id].bytes = 0;
+
+	/* The merge flow has received stats updates from firmware.
+	 * Distribute these stats to all subflows that form the merge.
+	 * The stats will collected from TC via the subflows.
+	 */
+	list_for_each_entry(link, &merge_flow->linked_flows, merge_flow.list) {
+		sub_flow = link->sub_flow.flow;
+		ctx_id = be32_to_cpu(sub_flow->meta.host_ctx_id);
+		priv->stats[ctx_id].pkts += pkts;
+		priv->stats[ctx_id].bytes += bytes;
+		max_t(u64, priv->stats[ctx_id].used, used);
+	}
+}
+
+static void
+nfp_flower_update_merge_stats(struct nfp_app *app,
+			      struct nfp_fl_payload *sub_flow)
+{
+	struct nfp_fl_payload_link *link;
+
+	/* Get merge flows that the subflow forms to distribute their stats. */
+	list_for_each_entry(link, &sub_flow->linked_flows, sub_flow.list)
+		__nfp_flower_update_merge_stats(app, link->merge_flow.flow);
+}
+
 /**
  * nfp_flower_get_stats() - Populates flow stats obtained from hardware.
  * @app:	Pointer to the APP handle
@@ -553,6 +1140,10 @@ nfp_flower_get_stats(struct nfp_app *app, struct net_device *netdev,
 	ctx_id = be32_to_cpu(nfp_flow->meta.host_ctx_id);
 
 	spin_lock_bh(&priv->stats_lock);
+	/* If request is for a sub_flow, update stats from merged flows. */
+	if (!list_empty(&nfp_flow->linked_flows))
+		nfp_flower_update_merge_stats(app, nfp_flow);
+
 	flow_stats_update(&flow->stats, priv->stats[ctx_id].bytes,
 			  priv->stats[ctx_id].pkts, priv->stats[ctx_id].used);
 
@@ -594,6 +1185,9 @@ static int nfp_flower_setup_tc_block_cb(enum tc_setup_type type,
 	case TC_SETUP_CLSFLOWER:
 		return nfp_flower_repr_offload(repr->app, repr->netdev,
 					       type_data);
+	case TC_SETUP_CLSMATCHALL:
+		return nfp_flower_setup_qos_offload(repr->app, repr->netdev,
+						    type_data);
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -603,10 +1197,14 @@ static int nfp_flower_setup_tc_block(struct net_device *netdev,
 				     struct tc_block_offload *f)
 {
 	struct nfp_repr *repr = netdev_priv(netdev);
+	struct nfp_flower_repr_priv *repr_priv;
 
 	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
 		return -EOPNOTSUPP;
 
+	repr_priv = repr->app_priv;
+	repr_priv->block_shared = tcf_block_shared(f->block);
+
 	switch (f->command) {
 	case TC_BLOCK_BIND:
 		return tcf_block_cb_register(f->block,
@@ -682,7 +1280,9 @@ nfp_flower_setup_indr_tc_block(struct net_device *netdev, struct nfp_app *app,
 	struct nfp_flower_priv *priv = app->priv;
 	int err;
 
-	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
+	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
+	    !(f->binder_type == TCF_BLOCK_BINDER_TYPE_CLSACT_EGRESS &&
+	      nfp_flower_internal_port_can_offload(app, netdev)))
 		return -EOPNOTSUPP;
 
 	switch (f->command) {