]> git.droids-corp.org - dpdk.git/commitdiff
net/bnxt: fix RSS RETA indirection table ops
authorAjit Khaparde <ajit.khaparde@broadcom.com>
Wed, 29 May 2019 21:02:24 +0000 (17:02 -0400)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 13 Jun 2019 14:37:59 +0000 (23:37 +0900)
We are trying to update the indirection table for all the VNICs.
We should update the table only for the default vnic0.

Fix the reta update function to only update table entries that are
selected by the update mask. Translate queue number to firmware
group ID when updating an entry.

Fix reta query op to only return table entries as identfied by the
provided mask. Translate firmware group IDs to queue numbers.

Removed extraneous code from bnxt_reta_query_op().

Fixes: d819382543f3 ("net/bnxt: add RSS redirection table operations")
Cc: stable@dpdk.org
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Lance Richardson <lance.richardson@broadcom.com>
Reviewed-by: Rahul Gupta <rahul.gupta@broadcom.com>
drivers/net/bnxt/bnxt_ethdev.c
drivers/net/bnxt/bnxt_hwrm.c

index 41771d8e25448ea5d936a3c6c1a3aa9461c9a88e..b5ed0536ce1c7438c7153639fc5fc1343488336f 100644 (file)
@@ -618,12 +618,6 @@ static void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
                        eth_dev->data->port_id);
 }
 
-static int bnxt_dev_lsc_intr_setup(struct rte_eth_dev *eth_dev)
-{
-       bnxt_print_link_info(eth_dev);
-       return 0;
-}
-
 /*
  * Determine whether the current configuration requires support for scattered
  * receive; return 1 if scattered receive is required and 0 if not.
@@ -977,30 +971,72 @@ static void bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
        bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
 }
 
+/* Return bnxt_rx_queue pointer corresponding to a given rxq. */
+static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
+{
+       if (qid >= bp->rx_nr_rings)
+               return NULL;
+
+       return bp->eth_dev->data->rx_queues[qid];
+}
+
+/* Return rxq corresponding to a given rss table ring/group ID. */
+static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
+{
+       unsigned int i;
+
+       for (i = 0; i < bp->rx_nr_rings; i++) {
+               if (bp->grp_info[i].fw_grp_id == fwr)
+                       return i;
+       }
+
+       return INVALID_HW_RING_ID;
+}
+
 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
                            struct rte_eth_rss_reta_entry64 *reta_conf,
                            uint16_t reta_size)
 {
        struct bnxt *bp = (struct bnxt *)eth_dev->data->dev_private;
        struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
-       struct bnxt_vnic_info *vnic;
+       struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
+       uint16_t tbl_size = HW_HASH_INDEX_SIZE;
+       uint16_t idx, sft;
        int i;
 
+       if (!vnic->rss_table)
+               return -EINVAL;
+
        if (!(dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
                return -EINVAL;
 
-       if (reta_size != HW_HASH_INDEX_SIZE) {
+       if (reta_size != tbl_size) {
                PMD_DRV_LOG(ERR, "The configured hash table lookup size "
                        "(%d) must equal the size supported by the hardware "
-                       "(%d)\n", reta_size, HW_HASH_INDEX_SIZE);
+                       "(%d)\n", reta_size, tbl_size);
                return -EINVAL;
        }
-       /* Update the RSS VNIC(s) */
-       for (i = 0; i < bp->max_vnics; i++) {
-               vnic = &bp->vnic_info[i];
-               memcpy(vnic->rss_table, reta_conf, reta_size);
-               bnxt_hwrm_vnic_rss_cfg(bp, vnic);
+
+       for (i = 0; i < reta_size; i++) {
+               struct bnxt_rx_queue *rxq;
+
+               idx = i / RTE_RETA_GROUP_SIZE;
+               sft = i % RTE_RETA_GROUP_SIZE;
+
+               if (!(reta_conf[idx].mask & (1ULL << sft)))
+                       continue;
+
+               rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
+               if (!rxq) {
+                       PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
+                       return -EINVAL;
+               }
+
+               vnic->rss_table[i] =
+                   vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
        }
+
+       bnxt_hwrm_vnic_rss_cfg(bp, vnic);
        return 0;
 }
 
@@ -1010,8 +1046,8 @@ static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
 {
        struct bnxt *bp = (struct bnxt *)eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
-       struct rte_intr_handle *intr_handle
-               = &bp->pdev->intr_handle;
+       uint16_t tbl_size = HW_HASH_INDEX_SIZE;
+       uint16_t idx, sft, i;
 
        /* Retrieve from the default VNIC */
        if (!vnic)
@@ -1019,18 +1055,28 @@ static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
        if (!vnic->rss_table)
                return -EINVAL;
 
-       if (reta_size != HW_HASH_INDEX_SIZE) {
+       if (reta_size != tbl_size) {
                PMD_DRV_LOG(ERR, "The configured hash table lookup size "
                        "(%d) must equal the size supported by the hardware "
-                       "(%d)\n", reta_size, HW_HASH_INDEX_SIZE);
+                       "(%d)\n", reta_size, tbl_size);
                return -EINVAL;
        }
-       /* EW - need to revisit here copying from uint64_t to uint16_t */
-       memcpy(reta_conf, vnic->rss_table, reta_size);
 
-       if (rte_intr_allow_others(intr_handle)) {
-               if (eth_dev->data->dev_conf.intr_conf.lsc != 0)
-                       bnxt_dev_lsc_intr_setup(eth_dev);
+       for (idx = 0, i = 0; i < reta_size; i++) {
+               idx = i / RTE_RETA_GROUP_SIZE;
+               sft = i % RTE_RETA_GROUP_SIZE;
+
+               if (reta_conf[idx].mask & (1ULL << sft)) {
+                       uint16_t qid;
+
+                       qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
+
+                       if (qid == INVALID_HW_RING_ID) {
+                               PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
+                               return -EINVAL;
+                       }
+                       reta_conf[idx].reta[sft] = qid;
+               }
        }
 
        return 0;
index c5aca89b213bfd34dbdfb554cb2517a4d55baf29..0c6af1622ea6aa40c1ca40b0b53d7f8494026f95 100644 (file)
@@ -1607,6 +1607,7 @@ int bnxt_hwrm_vnic_rss_cfg(struct bnxt *bp,
        req.hash_key_tbl_addr =
            rte_cpu_to_le_64(vnic->rss_hash_key_dma_addr);
        req.rss_ctx_idx = rte_cpu_to_le_16(vnic->rss_rule);
+       req.vnic_id = rte_cpu_to_le_16(vnic->fw_vnic_id);
 
        rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);