net/ice: support ether type filter on flow director
authorSimei Su <simei.su@intel.com>
Tue, 7 Jul 2020 03:21:15 +0000 (11:21 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 7 Jul 2020 21:38:28 +0000 (23:38 +0200)
This patch enables FDIR with input set ethertype.

Signed-off-by: Simei Su <simei.su@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
drivers/net/ice/ice_fdir_filter.c

index c3f23a0..c0f89db 100644 (file)
@@ -18,6 +18,9 @@
 
 #define ICE_FDIR_MAX_QREGION_SIZE      128
 
+#define ICE_FDIR_INSET_ETH (\
+       ICE_INSET_ETHERTYPE)
+
 #define ICE_FDIR_INSET_ETH_IPV4 (\
        ICE_INSET_DMAC | \
        ICE_INSET_IPV4_SRC | ICE_INSET_IPV4_DST | ICE_INSET_IPV4_TOS | \
@@ -102,6 +105,7 @@ static struct ice_pattern_match_item ice_fdir_pattern_os[] = {
 };
 
 static struct ice_pattern_match_item ice_fdir_pattern_comms[] = {
+       {pattern_ethertype,            ICE_FDIR_INSET_ETH,                   ICE_INSET_NONE},
        {pattern_eth_ipv4,             ICE_FDIR_INSET_ETH_IPV4,              ICE_INSET_NONE},
        {pattern_eth_ipv4_udp,         ICE_FDIR_INSET_ETH_IPV4_UDP,          ICE_INSET_NONE},
        {pattern_eth_ipv4_tcp,         ICE_FDIR_INSET_ETH_IPV4_TCP,          ICE_INSET_NONE},
@@ -894,6 +898,7 @@ ice_fdir_input_set_parse(uint64_t inset, enum ice_flow_field *field)
        };
        static const struct ice_inset_map ice_inset_map[] = {
                {ICE_INSET_DMAC, ICE_FLOW_FIELD_IDX_ETH_DA},
+               {ICE_INSET_ETHERTYPE, ICE_FLOW_FIELD_IDX_ETH_TYPE},
                {ICE_INSET_IPV4_SRC, ICE_FLOW_FIELD_IDX_IPV4_SA},
                {ICE_INSET_IPV4_DST, ICE_FLOW_FIELD_IDX_IPV4_DA},
                {ICE_INSET_IPV4_TOS, ICE_FLOW_FIELD_IDX_IPV4_DSCP},
@@ -1008,6 +1013,9 @@ ice_fdir_input_set_conf(struct ice_pf *pf, enum ice_fltr_ptype flow,
                else
                        PMD_DRV_LOG(ERR, "not supported tunnel type.");
                break;
+       case ICE_FLTR_PTYPE_NON_IP_L2:
+               ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_ETH_NON_IP);
+               break;
        default:
                PMD_DRV_LOG(ERR, "not supported filter type.");
                break;
@@ -1613,6 +1621,8 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
        };
        uint32_t vtc_flow_cpu;
 
+       enum rte_flow_item_type next_type;
+       uint16_t ether_type;
 
        for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
                if (item->last) {
@@ -1628,6 +1638,15 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                case RTE_FLOW_ITEM_TYPE_ETH:
                        eth_spec = item->spec;
                        eth_mask = item->mask;
+                       next_type = (item + 1)->type;
+
+                       if (next_type == RTE_FLOW_ITEM_TYPE_END &&
+                               (!eth_spec || !eth_mask)) {
+                               rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item, "NULL eth spec/mask.");
+                               return -rte_errno;
+                       }
 
                        if (eth_spec && eth_mask) {
                                if (!rte_is_zero_ether_addr(&eth_spec->src) ||
@@ -1639,18 +1658,33 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                        return -rte_errno;
                                }
 
-                               if (!rte_is_broadcast_ether_addr(&eth_mask->dst)) {
+                               if (rte_is_broadcast_ether_addr(&eth_mask->dst)) {
+                                       input_set |= ICE_INSET_DMAC;
+                                       rte_memcpy(&filter->input.ext_data.dst_mac,
+                                               &eth_spec->dst,
+                                               RTE_ETHER_ADDR_LEN);
+                               } else if (eth_mask->type == RTE_BE16(0xffff)) {
+                                       ether_type = rte_be_to_cpu_16(eth_spec->type);
+                                       if (ether_type == RTE_ETHER_TYPE_IPV4 ||
+                                               ether_type == RTE_ETHER_TYPE_IPV6) {
+                                               rte_flow_error_set(error, EINVAL,
+                                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                                       item,
+                                                       "Unsupported ether_type.");
+                                               return -rte_errno;
+                                       }
+
+                                       input_set |= ICE_INSET_ETHERTYPE;
+                                       rte_memcpy(&filter->input.ext_data.ether_type,
+                                               &eth_spec->type,
+                                               sizeof(eth_spec->type));
+                                       flow_type = ICE_FLTR_PTYPE_NON_IP_L2;
+                               } else {
                                        rte_flow_error_set(error, EINVAL,
                                                RTE_FLOW_ERROR_TYPE_ITEM,
                                                item,
-                                               "Invalid mac addr mask");
-                                       return -rte_errno;
+                                               "Invalid dst mac addr mask or ethertype mask");
                                }
-
-                               input_set |= ICE_INSET_DMAC;
-                               rte_memcpy(&filter->input.ext_data.dst_mac,
-                                          &eth_spec->dst,
-                                          RTE_ETHER_ADDR_LEN);
                        }
                        break;
                case RTE_FLOW_ITEM_TYPE_IPV4: