net/octeontx2: support flow parse actions
authorKiran Kumar K <kirankumark@marvell.com>
Sat, 29 Jun 2019 10:45:13 +0000 (16:15 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 4 Jul 2019 23:52:02 +0000 (01:52 +0200)
Adding support to parse flow actions like drop, count, mark, rss, queue.
On egress side, only drop and count actions were supported.

Signed-off-by: Kiran Kumar K <kirankumark@marvell.com>
Signed-off-by: Vivek Sharma <viveksharma@marvell.com>
drivers/net/octeontx2/otx2_flow_parse.c
drivers/net/octeontx2/otx2_rx.h

index b46fdd2..7f997ab 100644 (file)
@@ -679,3 +679,279 @@ otx2_flow_parse_la(struct otx2_parse_state *pst)
        /* Update pst if not validate only? clash check? */
        return otx2_flow_update_parse_state(pst, &info, lid, lt, 0);
 }
+
+static int
+parse_rss_action(struct rte_eth_dev *dev,
+                const struct rte_flow_attr *attr,
+                const struct rte_flow_action *act,
+                struct rte_flow_error *error)
+{
+       struct otx2_eth_dev *hw = dev->data->dev_private;
+       struct otx2_rss_info *rss_info = &hw->rss_info;
+       const struct rte_flow_action_rss *rss;
+       uint32_t i;
+
+       rss = (const struct rte_flow_action_rss *)act->conf;
+
+       /* Not supported */
+       if (attr->egress) {
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
+                                         attr, "No support of RSS in egress");
+       }
+
+       if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS)
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ACTION,
+                                         act, "multi-queue mode is disabled");
+
+       /* Parse RSS related parameters from configuration */
+       if (!rss || !rss->queue_num)
+               return rte_flow_error_set(error, EINVAL,
+                                         RTE_FLOW_ERROR_TYPE_ACTION,
+                                         act, "no valid queues");
+
+       if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT)
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ACTION, act,
+                                         "non-default RSS hash functions"
+                                         " are not supported");
+
+       if (rss->key_len && rss->key_len > RTE_DIM(rss_info->key))
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ACTION, act,
+                                         "RSS hash key too large");
+
+       if (rss->queue_num > rss_info->rss_size)
+               return rte_flow_error_set
+                       (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION, act,
+                        "too many queues for RSS context");
+
+       for (i = 0; i < rss->queue_num; i++) {
+               if (rss->queue[i] >= dev->data->nb_rx_queues)
+                       return rte_flow_error_set(error, EINVAL,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 act,
+                                                 "queue id > max number"
+                                                 " of queues");
+       }
+
+       return 0;
+}
+
+int
+otx2_flow_parse_actions(struct rte_eth_dev *dev,
+                       const struct rte_flow_attr *attr,
+                       const struct rte_flow_action actions[],
+                       struct rte_flow_error *error,
+                       struct rte_flow *flow)
+{
+       struct otx2_eth_dev *hw = dev->data->dev_private;
+       struct otx2_npc_flow_info *npc = &hw->npc_flow;
+       const struct rte_flow_action_count *act_count;
+       const struct rte_flow_action_mark *act_mark;
+       const struct rte_flow_action_queue *act_q;
+       const char *errmsg = NULL;
+       int sel_act, req_act = 0;
+       uint16_t pf_func;
+       int errcode = 0;
+       int mark = 0;
+       int rq = 0;
+
+       /* Initialize actions */
+       flow->ctr_id = NPC_COUNTER_NONE;
+
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+               otx2_npc_dbg("Action type = %d", actions->type);
+
+               switch (actions->type) {
+               case RTE_FLOW_ACTION_TYPE_VOID:
+                       break;
+               case RTE_FLOW_ACTION_TYPE_MARK:
+                       act_mark =
+                           (const struct rte_flow_action_mark *)actions->conf;
+
+                       /* We have only 16 bits. Use highest val for flag */
+                       if (act_mark->id > (OTX2_FLOW_FLAG_VAL - 2)) {
+                               errmsg = "mark value must be < 0xfffe";
+                               errcode = ENOTSUP;
+                               goto err_exit;
+                       }
+                       mark = act_mark->id + 1;
+                       req_act |= OTX2_FLOW_ACT_MARK;
+                       rte_atomic32_inc(&npc->mark_actions);
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_FLAG:
+                       mark = OTX2_FLOW_FLAG_VAL;
+                       req_act |= OTX2_FLOW_ACT_FLAG;
+                       rte_atomic32_inc(&npc->mark_actions);
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_COUNT:
+                       act_count =
+                               (const struct rte_flow_action_count *)
+                               actions->conf;
+
+                       if (act_count->shared == 1) {
+                               errmsg = "Shared Counters not supported";
+                               errcode = ENOTSUP;
+                               goto err_exit;
+                       }
+                       /* Indicates, need a counter */
+                       flow->ctr_id = 1;
+                       req_act |= OTX2_FLOW_ACT_COUNT;
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_DROP:
+                       req_act |= OTX2_FLOW_ACT_DROP;
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_QUEUE:
+                       /* Applicable only to ingress flow */
+                       act_q = (const struct rte_flow_action_queue *)
+                               actions->conf;
+                       rq = act_q->index;
+                       if (rq >= dev->data->nb_rx_queues) {
+                               errmsg = "invalid queue index";
+                               errcode = EINVAL;
+                               goto err_exit;
+                       }
+                       req_act |= OTX2_FLOW_ACT_QUEUE;
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_RSS:
+                       errcode = parse_rss_action(dev, attr, actions, error);
+                       if (errcode)
+                               return -rte_errno;
+
+                       req_act |= OTX2_FLOW_ACT_RSS;
+                       break;
+
+               case RTE_FLOW_ACTION_TYPE_SECURITY:
+                       /* Assumes user has already configured security
+                        * session for this flow. Associated conf is
+                        * opaque. When RTE security is implemented for otx2,
+                        * we need to verify that for specified security
+                        * session:
+                        *  action_type ==
+                        *    RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
+                        *  session_protocol ==
+                        *    RTE_SECURITY_PROTOCOL_IPSEC
+                        *
+                        * RSS is not supported with inline ipsec. Get the
+                        * rq from associated conf, or make
+                        * RTE_FLOW_ACTION_TYPE_QUEUE compulsory with this
+                        * action.
+                        * Currently, rq = 0 is assumed.
+                        */
+                       req_act |= OTX2_FLOW_ACT_SEC;
+                       rq = 0;
+                       break;
+               default:
+                       errmsg = "Unsupported action specified";
+                       errcode = ENOTSUP;
+                       goto err_exit;
+               }
+       }
+
+       /* Check if actions specified are compatible */
+       if (attr->egress) {
+               /* Only DROP/COUNT is supported */
+               if (!(req_act & OTX2_FLOW_ACT_DROP)) {
+                       errmsg = "DROP is required action for egress";
+                       errcode = EINVAL;
+                       goto err_exit;
+               } else if (req_act & ~(OTX2_FLOW_ACT_DROP |
+                                      OTX2_FLOW_ACT_COUNT)) {
+                       errmsg = "Unsupported action specified";
+                       errcode = ENOTSUP;
+                       goto err_exit;
+               }
+               flow->npc_action = NIX_TX_ACTIONOP_DROP;
+               goto set_pf_func;
+       }
+
+       /* We have already verified the attr, this is ingress.
+        * - Exactly one terminating action is supported
+        * - Exactly one of MARK or FLAG is supported
+        * - If terminating action is DROP, only count is valid.
+        */
+       sel_act = req_act & OTX2_FLOW_ACT_TERM;
+       if ((sel_act & (sel_act - 1)) != 0) {
+               errmsg = "Only one terminating action supported";
+               errcode = EINVAL;
+               goto err_exit;
+       }
+
+       if (req_act & OTX2_FLOW_ACT_DROP) {
+               sel_act = req_act & ~OTX2_FLOW_ACT_COUNT;
+               if ((sel_act & (sel_act - 1)) != 0) {
+                       errmsg = "Only COUNT action is supported "
+                               "with DROP ingress action";
+                       errcode = ENOTSUP;
+                       goto err_exit;
+               }
+       }
+
+       if ((req_act & (OTX2_FLOW_ACT_FLAG | OTX2_FLOW_ACT_MARK))
+           == (OTX2_FLOW_ACT_FLAG | OTX2_FLOW_ACT_MARK)) {
+               errmsg = "Only one of FLAG or MARK action is supported";
+               errcode = ENOTSUP;
+               goto err_exit;
+       }
+
+       /* Set NIX_RX_ACTIONOP */
+       if (req_act & OTX2_FLOW_ACT_DROP) {
+               flow->npc_action = NIX_RX_ACTIONOP_DROP;
+       } else if (req_act & OTX2_FLOW_ACT_QUEUE) {
+               flow->npc_action = NIX_RX_ACTIONOP_UCAST;
+               flow->npc_action |= (uint64_t)rq << 20;
+       } else if (req_act & OTX2_FLOW_ACT_RSS) {
+               /* When user added a rule for rss, first we will add the
+                *rule in MCAM and then update the action, once if we have
+                *FLOW_KEY_ALG index. So, till we update the action with
+                *flow_key_alg index, set the action to drop.
+                */
+               if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
+                       flow->npc_action = NIX_RX_ACTIONOP_DROP;
+               else
+                       flow->npc_action = NIX_RX_ACTIONOP_UCAST;
+       } else if (req_act & OTX2_FLOW_ACT_SEC) {
+               flow->npc_action = NIX_RX_ACTIONOP_UCAST_IPSEC;
+               flow->npc_action |= (uint64_t)rq << 20;
+       } else if (req_act & (OTX2_FLOW_ACT_FLAG | OTX2_FLOW_ACT_MARK)) {
+               flow->npc_action = NIX_RX_ACTIONOP_UCAST;
+       } else if (req_act & OTX2_FLOW_ACT_COUNT) {
+               /* Keep OTX2_FLOW_ACT_COUNT always at the end
+                * This is default action, when user specify only
+                * COUNT ACTION
+                */
+               flow->npc_action = NIX_RX_ACTIONOP_UCAST;
+       } else {
+               /* Should never reach here */
+               errmsg = "Invalid action specified";
+               errcode = EINVAL;
+               goto err_exit;
+       }
+
+       if (mark)
+               flow->npc_action |= (uint64_t)mark << 40;
+
+       if (rte_atomic32_read(&npc->mark_actions) == 1)
+               hw->rx_offload_flags |=
+                       NIX_RX_OFFLOAD_MARK_UPDATE_F;
+
+set_pf_func:
+       /* Ideally AF must ensure that correct pf_func is set */
+       pf_func = otx2_pfvf_func(hw->pf, hw->vf);
+       flow->npc_action |= (uint64_t)pf_func << 4;
+
+       return 0;
+
+err_exit:
+       rte_flow_error_set(error, errcode,
+                          RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
+                          errmsg);
+       return -rte_errno;
+}
index 0c3627c..db79451 100644 (file)
@@ -13,6 +13,7 @@
                                         sizeof(uint16_t))
 
 #define NIX_RX_OFFLOAD_PTYPE_F         BIT(1)
+#define NIX_RX_OFFLOAD_MARK_UPDATE_F   BIT(4)
 #define NIX_RX_OFFLOAD_TSTAMP_F        BIT(5)
 
 #define NIX_TIMESYNC_RX_OFFSET         8