net/octeontx2: add flow parsing for outer layers
authorKiran Kumar K <kirankumark@marvell.com>
Wed, 26 Jun 2019 10:34:22 +0000 (16:04 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 4 Jul 2019 23:52:02 +0000 (01:52 +0200)
Adding functionality to parse outer layers from ld to lh.
These will be used parse outer layers L2, L3, L4 and tunnel types.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Signed-off-by: Vivek Sharma <viveksharma@marvell.com>
drivers/net/octeontx2/Makefile
drivers/net/octeontx2/meson.build
drivers/net/octeontx2/otx2_flow_parse.c [new file with mode: 0644]

index 42429bf..636af65 100644 (file)
@@ -40,6 +40,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_OCTEONTX2_PMD) += \
        otx2_lookup.c   \
        otx2_ethdev.c   \
        otx2_flow_ctrl.c \
+       otx2_flow_parse.c \
        otx2_flow_utils.c \
        otx2_ethdev_irq.c \
        otx2_ethdev_ops.c \
index 75156dd..f608c49 100644 (file)
@@ -12,6 +12,7 @@ sources = files(
                'otx2_lookup.c',
                'otx2_ethdev.c',
                'otx2_flow_ctrl.c',
+               'otx2_flow_parse.c',
                'otx2_flow_utils.c',
                'otx2_ethdev_irq.c',
                'otx2_ethdev_ops.c',
diff --git a/drivers/net/octeontx2/otx2_flow_parse.c b/drivers/net/octeontx2/otx2_flow_parse.c
new file mode 100644 (file)
index 0000000..ed6c80f
--- /dev/null
@@ -0,0 +1,471 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(C) 2019 Marvell International Ltd.
+ */
+
+#include "otx2_ethdev.h"
+#include "otx2_flow.h"
+
+const struct rte_flow_item *
+otx2_flow_skip_void_and_any_items(const struct rte_flow_item *pattern)
+{
+       while ((pattern->type == RTE_FLOW_ITEM_TYPE_VOID) ||
+              (pattern->type == RTE_FLOW_ITEM_TYPE_ANY))
+               pattern++;
+
+       return pattern;
+}
+
+/*
+ * Tunnel+ESP, Tunnel+ICMP4/6, Tunnel+TCP, Tunnel+UDP,
+ * Tunnel+SCTP
+ */
+int
+otx2_flow_parse_lh(struct otx2_parse_state *pst)
+{
+       struct otx2_flow_item_info info;
+       char hw_mask[64];
+       int lid, lt;
+       int rc;
+
+       if (!pst->tunnel)
+               return 0;
+
+       info.hw_mask = &hw_mask;
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_hdr_len = 0;
+       lid = NPC_LID_LH;
+
+       switch (pst->pattern->type) {
+       case RTE_FLOW_ITEM_TYPE_UDP:
+               lt = NPC_LT_LH_TU_UDP;
+               info.def_mask = &rte_flow_item_udp_mask;
+               info.len = sizeof(struct rte_flow_item_udp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_TCP:
+               lt = NPC_LT_LH_TU_TCP;
+               info.def_mask = &rte_flow_item_tcp_mask;
+               info.len = sizeof(struct rte_flow_item_tcp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_SCTP:
+               lt = NPC_LT_LH_TU_SCTP;
+               info.def_mask = &rte_flow_item_sctp_mask;
+               info.len = sizeof(struct rte_flow_item_sctp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_ESP:
+               lt = NPC_LT_LH_TU_ESP;
+               info.def_mask = &rte_flow_item_esp_mask;
+               info.len = sizeof(struct rte_flow_item_esp);
+               break;
+       default:
+               return 0;
+       }
+
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       rc = otx2_flow_parse_item_basic(pst->pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, 0);
+}
+
+/* Tunnel+IPv4, Tunnel+IPv6 */
+int
+otx2_flow_parse_lg(struct otx2_parse_state *pst)
+{
+       struct otx2_flow_item_info info;
+       char hw_mask[64];
+       int lid, lt;
+       int rc;
+
+       if (!pst->tunnel)
+               return 0;
+
+       info.hw_mask = &hw_mask;
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_hdr_len = 0;
+       lid = NPC_LID_LG;
+
+       if (pst->pattern->type == RTE_FLOW_ITEM_TYPE_IPV4) {
+               lt = NPC_LT_LG_TU_IP;
+               info.def_mask = &rte_flow_item_ipv4_mask;
+               info.len = sizeof(struct rte_flow_item_ipv4);
+       } else if (pst->pattern->type == RTE_FLOW_ITEM_TYPE_IPV6) {
+               lt = NPC_LT_LG_TU_IP6;
+               info.def_mask = &rte_flow_item_ipv6_mask;
+               info.len = sizeof(struct rte_flow_item_ipv6);
+       } else {
+               /* There is no tunneled IP header */
+               return 0;
+       }
+
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       rc = otx2_flow_parse_item_basic(pst->pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, 0);
+}
+
+/* Tunnel+Ether */
+int
+otx2_flow_parse_lf(struct otx2_parse_state *pst)
+{
+       const struct rte_flow_item *pattern, *last_pattern;
+       struct rte_flow_item_eth hw_mask;
+       struct otx2_flow_item_info info;
+       int lid, lt, lflags;
+       int nr_vlans = 0;
+       int rc;
+
+       /* We hit this layer if there is a tunneling protocol */
+       if (!pst->tunnel)
+               return 0;
+
+       if (pst->pattern->type != RTE_FLOW_ITEM_TYPE_ETH)
+               return 0;
+
+       lid = NPC_LID_LF;
+       lt = NPC_LT_LF_TU_ETHER;
+       lflags = 0;
+
+       info.def_mask = &rte_flow_item_vlan_mask;
+       /* No match support for vlan tags */
+       info.hw_mask = NULL;
+       info.len = sizeof(struct rte_flow_item_vlan);
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_hdr_len = 0;
+
+       /* Look ahead and find out any VLAN tags. These can be
+        * detected but no data matching is available.
+        */
+       last_pattern = pst->pattern;
+       pattern = pst->pattern + 1;
+       pattern = otx2_flow_skip_void_and_any_items(pattern);
+       while (pattern->type == RTE_FLOW_ITEM_TYPE_VLAN) {
+               nr_vlans++;
+               rc = otx2_flow_parse_item_basic(pattern, &info, pst->error);
+               if (rc != 0)
+                       return rc;
+               last_pattern = pattern;
+               pattern++;
+               pattern = otx2_flow_skip_void_and_any_items(pattern);
+       }
+       otx2_npc_dbg("Nr_vlans = %d", nr_vlans);
+       switch (nr_vlans) {
+       case 0:
+               break;
+       case 1:
+               lflags = NPC_F_TU_ETHER_CTAG;
+               break;
+       case 2:
+               lflags = NPC_F_TU_ETHER_STAG_CTAG;
+               break;
+       default:
+               rte_flow_error_set(pst->error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_ITEM,
+                                  last_pattern,
+                                  "more than 2 vlans with tunneled Ethernet "
+                                  "not supported");
+               return -rte_errno;
+       }
+
+       info.def_mask = &rte_flow_item_eth_mask;
+       info.hw_mask = &hw_mask;
+       info.len = sizeof(struct rte_flow_item_eth);
+       info.hw_hdr_len = 0;
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       info.spec = NULL;
+       info.mask = NULL;
+
+       rc = otx2_flow_parse_item_basic(pst->pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       pst->pattern = last_pattern;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, lflags);
+}
+
+int
+otx2_flow_parse_le(struct otx2_parse_state *pst)
+{
+       /*
+        * We are positioned at UDP. Scan ahead and look for
+        * UDP encapsulated tunnel protocols. If available,
+        * parse them. In that case handle this:
+        *      - RTE spec assumes we point to tunnel header.
+        *      - NPC parser provides offset from UDP header.
+        */
+
+       /*
+        * Note: Add support to GENEVE, VXLAN_GPE when we
+        * upgrade DPDK
+        *
+        * Note: Better to split flags into two nibbles:
+        *      - Higher nibble can have flags
+        *      - Lower nibble to further enumerate protocols
+        *        and have flags based extraction
+        */
+       const struct rte_flow_item *pattern = pst->pattern;
+       struct otx2_flow_item_info info;
+       int lid, lt, lflags;
+       char hw_mask[64];
+       int rc;
+
+       if (pst->tunnel)
+               return 0;
+
+       if (pst->pattern->type == RTE_FLOW_ITEM_TYPE_MPLS)
+               return otx2_flow_parse_mpls(pst, NPC_LID_LE);
+
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_mask = NULL;
+       info.def_mask = NULL;
+       info.len = 0;
+       info.hw_hdr_len = 0;
+       lid = NPC_LID_LE;
+       lflags = 0;
+
+       /* Ensure we are not matching anything in UDP */
+       rc = otx2_flow_parse_item_basic(pattern, &info, pst->error);
+       if (rc)
+               return rc;
+
+       info.hw_mask = &hw_mask;
+       pattern = otx2_flow_skip_void_and_any_items(pattern);
+       otx2_npc_dbg("Pattern->type = %d", pattern->type);
+       switch (pattern->type) {
+       case RTE_FLOW_ITEM_TYPE_VXLAN:
+               lflags = NPC_F_UDP_VXLAN;
+               info.def_mask = &rte_flow_item_vxlan_mask;
+               info.len = sizeof(struct rte_flow_item_vxlan);
+               lt = NPC_LT_LE_VXLAN;
+               break;
+       case RTE_FLOW_ITEM_TYPE_GTPC:
+               lflags = NPC_F_UDP_GTP_GTPC;
+               info.def_mask = &rte_flow_item_gtp_mask;
+               info.len = sizeof(struct rte_flow_item_gtp);
+               lt = NPC_LT_LE_GTPC;
+               break;
+       case RTE_FLOW_ITEM_TYPE_GTPU:
+               lflags = NPC_F_UDP_GTP_GTPU_G_PDU;
+               info.def_mask = &rte_flow_item_gtp_mask;
+               info.len = sizeof(struct rte_flow_item_gtp);
+               lt = NPC_LT_LE_GTPU;
+               break;
+       case RTE_FLOW_ITEM_TYPE_GENEVE:
+               lflags = NPC_F_UDP_GENEVE;
+               info.def_mask = &rte_flow_item_geneve_mask;
+               info.len = sizeof(struct rte_flow_item_geneve);
+               lt = NPC_LT_LE_GENEVE;
+               break;
+       case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
+               lflags = NPC_F_UDP_VXLANGPE;
+               info.def_mask = &rte_flow_item_vxlan_gpe_mask;
+               info.len = sizeof(struct rte_flow_item_vxlan_gpe);
+               lt = NPC_LT_LE_VXLANGPE;
+               break;
+       default:
+               return 0;
+       }
+
+       pst->tunnel = 1;
+
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       rc = otx2_flow_parse_item_basic(pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, lflags);
+}
+
+static int
+flow_parse_mpls_label_stack(struct otx2_parse_state *pst, int *flag)
+{
+       int nr_labels = 0;
+       const struct rte_flow_item *pattern = pst->pattern;
+       struct otx2_flow_item_info info;
+       int rc;
+       uint8_t flag_list[] = {0, NPC_F_MPLS_2_LABELS,
+               NPC_F_MPLS_3_LABELS, NPC_F_MPLS_4_LABELS};
+
+       /*
+        * pst->pattern points to first MPLS label. We only check
+        * that subsequent labels do not have anything to match.
+        */
+       info.def_mask = &rte_flow_item_mpls_mask;
+       info.hw_mask = NULL;
+       info.len = sizeof(struct rte_flow_item_mpls);
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_hdr_len = 0;
+
+       while (pattern->type == RTE_FLOW_ITEM_TYPE_MPLS) {
+               nr_labels++;
+
+               /* Basic validation of 2nd/3rd/4th mpls item */
+               if (nr_labels > 1) {
+                       rc = otx2_flow_parse_item_basic(pattern, &info,
+                                                       pst->error);
+                       if (rc != 0)
+                               return rc;
+               }
+               pst->last_pattern = pattern;
+               pattern++;
+               pattern = otx2_flow_skip_void_and_any_items(pattern);
+       }
+
+       if (nr_labels > 4) {
+               rte_flow_error_set(pst->error, ENOTSUP,
+                                  RTE_FLOW_ERROR_TYPE_ITEM,
+                                  pst->last_pattern,
+                                  "more than 4 mpls labels not supported");
+               return -rte_errno;
+       }
+
+       *flag = flag_list[nr_labels - 1];
+       return 0;
+}
+
+int
+otx2_flow_parse_mpls(struct otx2_parse_state *pst, int lid)
+{
+       /* Find number of MPLS labels */
+       struct rte_flow_item_mpls hw_mask;
+       struct otx2_flow_item_info info;
+       int lt, lflags;
+       int rc;
+
+       lflags = 0;
+
+       if (lid == NPC_LID_LC)
+               lt = NPC_LT_LC_MPLS;
+       else if (lid == NPC_LID_LD)
+               lt = NPC_LT_LD_TU_MPLS_IN_IP;
+       else
+               lt = NPC_LT_LE_TU_MPLS_IN_UDP;
+
+       /* Prepare for parsing the first item */
+       info.def_mask = &rte_flow_item_mpls_mask;
+       info.hw_mask = &hw_mask;
+       info.len = sizeof(struct rte_flow_item_mpls);
+       info.spec = NULL;
+       info.mask = NULL;
+       info.hw_hdr_len = 0;
+
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       rc = otx2_flow_parse_item_basic(pst->pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       /*
+        * Parse for more labels.
+        * This sets lflags and pst->last_pattern correctly.
+        */
+       rc = flow_parse_mpls_label_stack(pst, &lflags);
+       if (rc != 0)
+               return rc;
+
+       pst->tunnel = 1;
+       pst->pattern = pst->last_pattern;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, lflags);
+}
+
+/*
+ * ICMP, ICMP6, UDP, TCP, SCTP, VXLAN, GRE, NVGRE,
+ * GTP, GTPC, GTPU, ESP
+ *
+ * Note: UDP tunnel protocols are identified by flags.
+ *       LPTR for these protocol still points to UDP
+ *       header. Need flag based extraction to support
+ *       this.
+ */
+int
+otx2_flow_parse_ld(struct otx2_parse_state *pst)
+{
+       char hw_mask[NPC_MAX_EXTRACT_DATA_LEN];
+       struct otx2_flow_item_info info;
+       int lid, lt, lflags;
+       int rc;
+
+       if (pst->tunnel) {
+               /* We have already parsed MPLS or IPv4/v6 followed
+                * by MPLS or IPv4/v6. Subsequent TCP/UDP etc
+                * would be parsed as tunneled versions. Skip
+                * this layer, except for tunneled MPLS. If LC is
+                * MPLS, we have anyway skipped all stacked MPLS
+                * labels.
+                */
+               if (pst->pattern->type == RTE_FLOW_ITEM_TYPE_MPLS)
+                       return otx2_flow_parse_mpls(pst, NPC_LID_LD);
+               return 0;
+       }
+       info.hw_mask = &hw_mask;
+       info.spec = NULL;
+       info.mask = NULL;
+       info.def_mask = NULL;
+       info.len = 0;
+       info.hw_hdr_len = 0;
+
+       lid = NPC_LID_LD;
+       lflags = 0;
+
+       otx2_npc_dbg("Pst->pattern->type = %d", pst->pattern->type);
+       switch (pst->pattern->type) {
+       case RTE_FLOW_ITEM_TYPE_ICMP:
+               if (pst->lt[NPC_LID_LC] == NPC_LT_LC_IP6)
+                       lt = NPC_LT_LD_ICMP6;
+               else
+                       lt = NPC_LT_LD_ICMP;
+               info.def_mask = &rte_flow_item_icmp_mask;
+               info.len = sizeof(struct rte_flow_item_icmp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_UDP:
+               lt = NPC_LT_LD_UDP;
+               info.def_mask = &rte_flow_item_udp_mask;
+               info.len = sizeof(struct rte_flow_item_udp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_TCP:
+               lt = NPC_LT_LD_TCP;
+               info.def_mask = &rte_flow_item_tcp_mask;
+               info.len = sizeof(struct rte_flow_item_tcp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_SCTP:
+               lt = NPC_LT_LD_SCTP;
+               info.def_mask = &rte_flow_item_sctp_mask;
+               info.len = sizeof(struct rte_flow_item_sctp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_ESP:
+               lt = NPC_LT_LD_ESP;
+               info.def_mask = &rte_flow_item_esp_mask;
+               info.len = sizeof(struct rte_flow_item_esp);
+               break;
+       case RTE_FLOW_ITEM_TYPE_GRE:
+               lt = NPC_LT_LD_GRE;
+               info.def_mask = &rte_flow_item_gre_mask;
+               info.len = sizeof(struct rte_flow_item_gre);
+               break;
+       case RTE_FLOW_ITEM_TYPE_NVGRE:
+               lt = NPC_LT_LD_GRE;
+               lflags = NPC_F_GRE_NVGRE;
+               info.def_mask = &rte_flow_item_nvgre_mask;
+               info.len = sizeof(struct rte_flow_item_nvgre);
+               /* Further IP/Ethernet are parsed as tunneled */
+               pst->tunnel = 1;
+               break;
+       default:
+               return 0;
+       }
+
+       otx2_flow_get_hw_supp_mask(pst, &info, lid, lt);
+       rc = otx2_flow_parse_item_basic(pst->pattern, &info, pst->error);
+       if (rc != 0)
+               return rc;
+
+       return otx2_flow_update_parse_state(pst, &info, lid, lt, lflags);
+}