net/ice: fix memory leak when releasing VSI
[dpdk.git] / drivers / net / ice / ice_fdir_filter.c
index 6342b56..745d729 100644 (file)
 
 #define ICE_FDIR_MAX_QREGION_SIZE      128
 
+#define ICE_FDIR_INSET_ETH (\
+       ICE_INSET_DMAC | ICE_INSET_SMAC | ICE_INSET_ETHERTYPE)
+
 #define ICE_FDIR_INSET_ETH_IPV4 (\
-       ICE_INSET_DMAC | \
+       ICE_FDIR_INSET_ETH | \
        ICE_INSET_IPV4_SRC | ICE_INSET_IPV4_DST | ICE_INSET_IPV4_TOS | \
        ICE_INSET_IPV4_TTL | ICE_INSET_IPV4_PROTO)
 
        ICE_FDIR_INSET_VXLAN_IPV4 | \
        ICE_INSET_TUN_SCTP_SRC_PORT | ICE_INSET_TUN_SCTP_DST_PORT)
 
-#define ICE_FDIR_INSET_GTPU (\
+#define ICE_FDIR_INSET_IPV4_GTPU (\
        ICE_INSET_IPV4_SRC | ICE_INSET_IPV4_DST | ICE_INSET_GTPU_TEID)
 
-#define ICE_FDIR_INSET_GTPU_EH (\
+#define ICE_FDIR_INSET_IPV4_GTPU_EH (\
        ICE_INSET_IPV4_SRC | ICE_INSET_IPV4_DST | \
        ICE_INSET_GTPU_TEID | ICE_INSET_GTPU_QFI)
 
+#define ICE_FDIR_INSET_IPV6_GTPU (\
+       ICE_INSET_IPV6_SRC | ICE_INSET_IPV6_DST | ICE_INSET_GTPU_TEID)
+
+#define ICE_FDIR_INSET_IPV6_GTPU_EH (\
+       ICE_INSET_IPV6_SRC | ICE_INSET_IPV6_DST | \
+       ICE_INSET_GTPU_TEID | ICE_INSET_GTPU_QFI)
+
 static struct ice_pattern_match_item ice_fdir_pattern_os[] = {
        {pattern_eth_ipv4,             ICE_FDIR_INSET_ETH_IPV4,              ICE_INSET_NONE},
        {pattern_eth_ipv4_udp,         ICE_FDIR_INSET_ETH_IPV4_UDP,          ICE_INSET_NONE},
@@ -102,6 +112,7 @@ 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},
@@ -126,8 +137,10 @@ static struct ice_pattern_match_item ice_fdir_pattern_comms[] = {
                                       ICE_FDIR_INSET_VXLAN_IPV4_TCP,        ICE_INSET_NONE},
        {pattern_eth_ipv4_udp_vxlan_eth_ipv4_sctp,
                                       ICE_FDIR_INSET_VXLAN_IPV4_SCTP,       ICE_INSET_NONE},
-       {pattern_eth_ipv4_gtpu,        ICE_FDIR_INSET_GTPU,                  ICE_INSET_NONE},
-       {pattern_eth_ipv4_gtpu_eh,     ICE_FDIR_INSET_GTPU_EH,               ICE_INSET_NONE},
+       {pattern_eth_ipv4_gtpu,        ICE_FDIR_INSET_IPV4_GTPU,             ICE_INSET_NONE},
+       {pattern_eth_ipv4_gtpu_eh,     ICE_FDIR_INSET_IPV4_GTPU_EH,          ICE_INSET_NONE},
+       {pattern_eth_ipv6_gtpu,        ICE_FDIR_INSET_IPV6_GTPU,             ICE_INSET_NONE},
+       {pattern_eth_ipv6_gtpu_eh,     ICE_FDIR_INSET_IPV6_GTPU_EH,          ICE_INSET_NONE},
 };
 
 static struct ice_flow_parser ice_fdir_parser_os;
@@ -584,7 +597,7 @@ ice_fdir_prof_rm(struct ice_pf *pf, enum ice_fltr_ptype ptype, bool is_tunnel)
                                                     hw_prof->vsi_h[i]);
                        ice_rem_prof_id_flow(hw, ICE_BLK_FD,
                                             vsi_num, ptype);
-                       ice_flow_rem_entry(hw,
+                       ice_flow_rem_entry(hw, ICE_BLK_FD,
                                           hw_prof->entry_h[i][is_tunnel]);
                        hw_prof->entry_h[i][is_tunnel] = 0;
                }
@@ -778,6 +791,21 @@ ice_fdir_cross_prof_conflict(struct ice_pf *pf,
                        (pf, cflct_ptype, is_tunnel))
                        goto err;
                break;
+       /* IPv6 GTPU */
+       case ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER:
+               cflct_ptype = ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER;
+               if (!ice_fdir_prof_resolve_conflict
+                       (pf, cflct_ptype, is_tunnel))
+                       goto err;
+               cflct_ptype = ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER;
+               if (!ice_fdir_prof_resolve_conflict
+                       (pf, cflct_ptype, is_tunnel))
+                       goto err;
+               cflct_ptype = ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER;
+               if (!ice_fdir_prof_resolve_conflict
+                       (pf, cflct_ptype, is_tunnel))
+                       goto err;
+               break;
        /* IPv6 */
        case ICE_FLTR_PTYPE_NONF_IPV6_UDP:
        case ICE_FLTR_PTYPE_NONF_IPV6_TCP:
@@ -876,7 +904,7 @@ ice_fdir_hw_tbl_conf(struct ice_pf *pf, struct ice_vsi *vsi,
 err_add_entry:
        vsi_num = ice_get_hw_vsi_num(hw, vsi->idx);
        ice_rem_prof_id_flow(hw, ICE_BLK_FD, vsi_num, prof_id);
-       ice_flow_rem_entry(hw, entry_1);
+       ice_flow_rem_entry(hw, ICE_BLK_FD, entry_1);
 err_add_prof:
        ice_flow_rem_prof(hw, ICE_BLK_FD, prof_id);
 
@@ -894,6 +922,7 @@ 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},
@@ -956,33 +985,41 @@ ice_fdir_input_set_conf(struct ice_pf *pf, enum ice_fltr_ptype flow,
        switch (flow) {
        case ICE_FLTR_PTYPE_NONF_IPV4_UDP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_UDP |
-                                 ICE_FLOW_SEG_HDR_IPV4);
+                                 ICE_FLOW_SEG_HDR_IPV4 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV4_TCP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_TCP |
-                                 ICE_FLOW_SEG_HDR_IPV4);
+                                 ICE_FLOW_SEG_HDR_IPV4 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV4_SCTP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_SCTP |
-                                 ICE_FLOW_SEG_HDR_IPV4);
+                                 ICE_FLOW_SEG_HDR_IPV4 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV4_OTHER:
-               ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_IPV4);
+               ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_IPV4 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV6_UDP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_UDP |
-                                 ICE_FLOW_SEG_HDR_IPV6);
+                                 ICE_FLOW_SEG_HDR_IPV6 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV6_TCP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_TCP |
-                                 ICE_FLOW_SEG_HDR_IPV6);
+                                 ICE_FLOW_SEG_HDR_IPV6 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV6_SCTP:
                ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_SCTP |
-                                 ICE_FLOW_SEG_HDR_IPV6);
+                                 ICE_FLOW_SEG_HDR_IPV6 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV6_OTHER:
-               ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_IPV6);
+               ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_IPV6 |
+                                 ICE_FLOW_SEG_HDR_IPV_OTHER);
                break;
        case ICE_FLTR_PTYPE_NONF_IPV4_GTPU_IPV4_UDP:
        case ICE_FLTR_PTYPE_NONF_IPV4_GTPU_IPV4_TCP:
@@ -990,14 +1027,32 @@ ice_fdir_input_set_conf(struct ice_pf *pf, enum ice_fltr_ptype flow,
        case ICE_FLTR_PTYPE_NONF_IPV4_GTPU_IPV4_OTHER:
                if (ttype == ICE_FDIR_TUNNEL_TYPE_GTPU)
                        ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_GTPU_IP |
-                                         ICE_FLOW_SEG_HDR_IPV4);
+                                         ICE_FLOW_SEG_HDR_IPV4 |
+                                         ICE_FLOW_SEG_HDR_IPV_OTHER);
                else if (ttype == ICE_FDIR_TUNNEL_TYPE_GTPU_EH)
                        ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_GTPU_EH |
                                          ICE_FLOW_SEG_HDR_GTPU_IP |
-                                         ICE_FLOW_SEG_HDR_IPV4);
+                                         ICE_FLOW_SEG_HDR_IPV4 |
+                                         ICE_FLOW_SEG_HDR_IPV_OTHER);
                else
                        PMD_DRV_LOG(ERR, "not supported tunnel type.");
                break;
+       case ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER:
+               if (ttype == ICE_FDIR_TUNNEL_TYPE_GTPU)
+                       ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_GTPU_IP |
+                                         ICE_FLOW_SEG_HDR_IPV6 |
+                                         ICE_FLOW_SEG_HDR_IPV_OTHER);
+               else if (ttype == ICE_FDIR_TUNNEL_TYPE_GTPU_EH)
+                       ICE_FLOW_SET_HDRS(seg, ICE_FLOW_SEG_HDR_GTPU_EH |
+                                         ICE_FLOW_SEG_HDR_GTPU_IP |
+                                         ICE_FLOW_SEG_HDR_IPV6 |
+                                         ICE_FLOW_SEG_HDR_IPV_OTHER);
+               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;
@@ -1061,6 +1116,9 @@ ice_fdir_init(struct ice_adapter *ad)
        struct ice_flow_parser *parser;
        int ret;
 
+       if (ad->hw.dcf_enabled)
+               return 0;
+
        ret = ice_fdir_setup(pf);
        if (ret)
                return ret;
@@ -1081,6 +1139,9 @@ ice_fdir_uninit(struct ice_adapter *ad)
        struct ice_pf *pf = &ad->pf;
        struct ice_flow_parser *parser;
 
+       if (ad->hw.dcf_enabled)
+               return;
+
        if (ad->active_pkg_type == ICE_PKG_TYPE_COMMS)
                parser = &ice_fdir_parser_comms;
        else
@@ -1114,6 +1175,7 @@ ice_fdir_add_del_filter(struct ice_pf *pf,
        filter->input.dest_vsi = pf->main_vsi->idx;
 
        memset(&desc, 0, sizeof(desc));
+       filter->input.comp_report = ICE_FXD_FLTR_QW0_COMP_REPORT_SW;
        ice_fdir_get_prgm_desc(hw, &filter->input, &desc, add);
 
        is_tun = ice_fdir_is_tunnel_profile(filter->tunnel_type);
@@ -1533,7 +1595,7 @@ ice_fdir_parse_action(struct ice_adapter *ad,
                }
        }
 
-       if (dest_num == 0 || dest_num >= 2) {
+       if (dest_num >= 2) {
                rte_flow_error_set(error, EINVAL,
                           RTE_FLOW_ERROR_TYPE_ACTION, actions,
                           "Unsupported action combination");
@@ -1554,6 +1616,18 @@ ice_fdir_parse_action(struct ice_adapter *ad,
                return -rte_errno;
        }
 
+       if (dest_num + mark_num + counter_num == 0) {
+               rte_flow_error_set(error, EINVAL,
+                          RTE_FLOW_ERROR_TYPE_ACTION, actions,
+                          "Empty action");
+               return -rte_errno;
+       }
+
+       /* set default action to PASSTHRU mode, in "mark/count only" case. */
+       if (dest_num == 0)
+               filter->input.dest_ctl =
+                       ICE_FLTR_PRGM_DESC_DEST_DIRECT_PKT_OTHER;
+
        return 0;
 }
 
@@ -1583,7 +1657,8 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF
        };
        uint32_t vtc_flow_cpu;
-
+       uint16_t ether_type;
+       enum rte_flow_item_type next_type;
 
        for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
                if (item->last) {
@@ -1599,29 +1674,43 @@ 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 (eth_spec && eth_mask) {
-                               if (!rte_is_zero_ether_addr(&eth_spec->src) ||
-                                   !rte_is_zero_ether_addr(&eth_mask->src)) {
-                                       rte_flow_error_set(error, EINVAL,
-                                               RTE_FLOW_ERROR_TYPE_ITEM,
-                                               item,
-                                               "Src mac not support");
-                                       return -rte_errno;
+                               if (!rte_is_zero_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);
                                }
 
-                               if (!rte_is_broadcast_ether_addr(&eth_mask->dst)) {
-                                       rte_flow_error_set(error, EINVAL,
-                                               RTE_FLOW_ERROR_TYPE_ITEM,
-                                               item,
-                                               "Invalid mac addr mask");
-                                       return -rte_errno;
+                               if (!rte_is_zero_ether_addr(&eth_mask->src)) {
+                                       input_set |= ICE_INSET_SMAC;
+                                       rte_memcpy(&filter->input.ext_data.src_mac,
+                                                  &eth_spec->src,
+                                                  RTE_ETHER_ADDR_LEN);
                                }
 
-                               input_set |= ICE_INSET_DMAC;
-                               rte_memcpy(&filter->input.ext_data.dst_mac,
-                                          &eth_spec->dst,
-                                          RTE_ETHER_ADDR_LEN);
+                               /* Ignore this field except for ICE_FLTR_PTYPE_NON_IP_L2 */
+                               if (eth_mask->type == RTE_BE16(0xffff) &&
+                                   next_type == RTE_FLOW_ITEM_TYPE_END) {
+                                       input_set |= ICE_INSET_ETHERTYPE;
+                                       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;
+                                       }
+
+                                       rte_memcpy(&filter->input.ext_data.ether_type,
+                                                  &eth_spec->type,
+                                                  sizeof(eth_spec->type));
+                                       flow_type = ICE_FLTR_PTYPE_NON_IP_L2;
+                               }
                        }
                        break;
                case RTE_FLOW_ITEM_TYPE_IPV4:
@@ -1658,9 +1747,9 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                        input_set |= ICE_INSET_IPV4_PROTO;
 
                                filter->input.ip.v4.dst_ip =
-                                       ipv4_spec->hdr.src_addr;
-                               filter->input.ip.v4.src_ip =
                                        ipv4_spec->hdr.dst_addr;
+                               filter->input.ip.v4.src_ip =
+                                       ipv4_spec->hdr.src_addr;
                                filter->input.ip.v4.tos =
                                        ipv4_spec->hdr.type_of_service;
                                filter->input.ip.v4.ttl =
@@ -1705,9 +1794,9 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                        input_set |= ICE_INSET_IPV6_HOP_LIMIT;
 
                                rte_memcpy(filter->input.ip.v6.dst_ip,
-                                          ipv6_spec->hdr.src_addr, 16);
-                               rte_memcpy(filter->input.ip.v6.src_ip,
                                           ipv6_spec->hdr.dst_addr, 16);
+                               rte_memcpy(filter->input.ip.v6.src_ip,
+                                          ipv6_spec->hdr.src_addr, 16);
 
                                vtc_flow_cpu =
                                      rte_be_to_cpu_32(ipv6_spec->hdr.vtc_flow);
@@ -1759,14 +1848,14 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                /* Get filter info */
                                if (l3 == RTE_FLOW_ITEM_TYPE_IPV4) {
                                        filter->input.ip.v4.dst_port =
-                                               tcp_spec->hdr.src_port;
-                                       filter->input.ip.v4.src_port =
                                                tcp_spec->hdr.dst_port;
+                                       filter->input.ip.v4.src_port =
+                                               tcp_spec->hdr.src_port;
                                } else if (l3 == RTE_FLOW_ITEM_TYPE_IPV6) {
                                        filter->input.ip.v6.dst_port =
-                                               tcp_spec->hdr.src_port;
-                                       filter->input.ip.v6.src_port =
                                                tcp_spec->hdr.dst_port;
+                                       filter->input.ip.v6.src_port =
+                                               tcp_spec->hdr.src_port;
                                }
                        }
                        break;
@@ -1802,14 +1891,14 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                /* Get filter info */
                                if (l3 == RTE_FLOW_ITEM_TYPE_IPV4) {
                                        filter->input.ip.v4.dst_port =
-                                               udp_spec->hdr.src_port;
-                                       filter->input.ip.v4.src_port =
                                                udp_spec->hdr.dst_port;
+                                       filter->input.ip.v4.src_port =
+                                               udp_spec->hdr.src_port;
                                } else if (l3 == RTE_FLOW_ITEM_TYPE_IPV6) {
                                        filter->input.ip.v6.src_port =
-                                               udp_spec->hdr.dst_port;
-                                       filter->input.ip.v6.dst_port =
                                                udp_spec->hdr.src_port;
+                                       filter->input.ip.v6.dst_port =
+                                               udp_spec->hdr.dst_port;
                                }
                        }
                        break;
@@ -1844,14 +1933,14 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
                                /* Get filter info */
                                if (l3 == RTE_FLOW_ITEM_TYPE_IPV4) {
                                        filter->input.ip.v4.dst_port =
-                                               sctp_spec->hdr.src_port;
-                                       filter->input.ip.v4.src_port =
                                                sctp_spec->hdr.dst_port;
+                                       filter->input.ip.v4.src_port =
+                                               sctp_spec->hdr.src_port;
                                } else if (l3 == RTE_FLOW_ITEM_TYPE_IPV6) {
                                        filter->input.ip.v6.dst_port =
-                                               sctp_spec->hdr.src_port;
-                                       filter->input.ip.v6.src_port =
                                                sctp_spec->hdr.dst_port;
+                                       filter->input.ip.v6.src_port =
+                                               sctp_spec->hdr.src_port;
                                }
                        }
                        break;
@@ -1919,8 +2008,12 @@ ice_fdir_parse_pattern(__rte_unused struct ice_adapter *ad,
        }
 
        if (tunnel_type == ICE_FDIR_TUNNEL_TYPE_GTPU ||
-           tunnel_type == ICE_FDIR_TUNNEL_TYPE_GTPU_EH)
-               flow_type = ICE_FLTR_PTYPE_NONF_IPV4_GTPU_IPV4_OTHER;
+           tunnel_type == ICE_FDIR_TUNNEL_TYPE_GTPU_EH) {
+               if (flow_type == ICE_FLTR_PTYPE_NONF_IPV4_UDP)
+                       flow_type = ICE_FLTR_PTYPE_NONF_IPV4_GTPU_IPV4_OTHER;
+               else
+                       flow_type = ICE_FLTR_PTYPE_NONF_IPV6_GTPU_IPV6_OTHER;
+       }
 
        filter->tunnel_type = tunnel_type;
        filter->input.flow_type = flow_type;