net/bnxt: create default flow rules for port reprentor
authorSomnath Kotur <somnath.kotur@broadcom.com>
Thu, 2 Jul 2020 23:28:33 +0000 (16:28 -0700)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 7 Jul 2020 21:38:27 +0000 (23:38 +0200)
Invoked 3 new APIs for the default flow create/destroy and to get
the action ptr for a default flow.
Changed ulp_intf_update() to accept rte_eth_dev as input and invoke
the same from the VF rep start function.
ULP Mark Manager will indicate if the cfa_code returned in the
Rx completion descriptor was for one of the default flow rules
created for the VF representor conduit. The mark_id returned
in such a case would be the VF rep's DPDK Port id, which can be
used to get the corresponding rte_eth_dev struct in bnxt_vf_recv

Signed-off-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Venkat Duvvuru <venkatkumar.duvvuru@broadcom.com>
Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
drivers/net/bnxt/bnxt.h
drivers/net/bnxt/bnxt_reps.c
drivers/net/bnxt/bnxt_reps.h
drivers/net/bnxt/bnxt_rxr.c
drivers/net/bnxt/bnxt_txq.h

index 32acced..f16bf33 100644 (file)
@@ -806,8 +806,10 @@ struct bnxt_vf_representor {
        uint16_t                fw_fid;
        uint16_t                dflt_vnic_id;
        uint16_t                svif;
-       uint16_t                tx_cfa_action;
+       uint32_t                vfr_tx_cfa_action;
        uint16_t                rx_cfa_code;
+       uint32_t                rep2vf_flow_id;
+       uint32_t                vf2rep_flow_id;
        /* Private data store of associated PF/Trusted VF */
        struct rte_eth_dev      *parent_dev;
        uint8_t                 mac_addr[RTE_ETHER_ADDR_LEN];
index bdbce6e..47e2dd7 100644 (file)
@@ -12,6 +12,9 @@
 #include "bnxt_txr.h"
 #include "bnxt_hwrm.h"
 #include "hsi_struct_def_dpdk.h"
+#include "bnxt_tf_common.h"
+#include "ulp_port_db.h"
+#include "ulp_flow_db.h"
 
 static const struct eth_dev_ops bnxt_vf_rep_dev_ops = {
        .dev_infos_get = bnxt_vf_rep_dev_info_get_op,
@@ -29,30 +32,20 @@ static const struct eth_dev_ops bnxt_vf_rep_dev_ops = {
 };
 
 uint16_t
-bnxt_vfr_recv(struct bnxt *bp, uint16_t cfa_code, uint16_t queue_id,
-             struct rte_mbuf *mbuf)
+bnxt_vfr_recv(uint16_t port_id, uint16_t queue_id, struct rte_mbuf *mbuf)
 {
        struct bnxt_sw_rx_bd *prod_rx_buf;
        struct bnxt_rx_ring_info *rep_rxr;
        struct bnxt_rx_queue *rep_rxq;
        struct rte_eth_dev *vfr_eth_dev;
        struct bnxt_vf_representor *vfr_bp;
-       uint16_t vf_id;
        uint16_t mask;
        uint8_t que;
 
-       vf_id = bp->cfa_code_map[cfa_code];
-       /* cfa_code is invalid OR vf_id > MAX REP. Assume normal Rx */
-       if (vf_id == BNXT_VF_IDX_INVALID || vf_id > BNXT_MAX_VF_REPS)
-               return 1;
-       vfr_eth_dev = bp->rep_info[vf_id].vfr_eth_dev;
+       vfr_eth_dev = &rte_eth_devices[port_id];
        if (!vfr_eth_dev)
                return 1;
        vfr_bp = vfr_eth_dev->data->dev_private;
-       if (vfr_bp->rx_cfa_code != cfa_code) {
-               /* cfa_code not meant for this VF rep!!?? */
-               return 1;
-       }
        /* If rxq_id happens to be > max rep_queue, use rxq0 */
        que = queue_id < BNXT_MAX_VF_REP_RINGS ? queue_id : 0;
        rep_rxq = vfr_bp->rx_queues[que];
@@ -127,7 +120,7 @@ bnxt_vf_rep_tx_burst(void *tx_queue,
        pthread_mutex_lock(&parent->rep_info->vfr_lock);
        ptxq = parent->tx_queues[qid];
 
-       ptxq->tx_cfa_action = vf_rep_bp->tx_cfa_action;
+       ptxq->vfr_tx_cfa_action = vf_rep_bp->vfr_tx_cfa_action;
 
        for (i = 0; i < nb_pkts; i++) {
                vf_rep_bp->tx_bytes[qid] += tx_pkts[i]->pkt_len;
@@ -135,7 +128,7 @@ bnxt_vf_rep_tx_burst(void *tx_queue,
        }
 
        rc = bnxt_xmit_pkts(ptxq, tx_pkts, nb_pkts);
-       ptxq->tx_cfa_action = 0;
+       ptxq->vfr_tx_cfa_action = 0;
        pthread_mutex_unlock(&parent->rep_info->vfr_lock);
 
        return rc;
@@ -252,10 +245,67 @@ int bnxt_vf_rep_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_compl)
        return rc;
 }
 
-static int bnxt_vfr_alloc(struct bnxt_vf_representor *vfr)
+static int bnxt_tf_vfr_alloc(struct rte_eth_dev *vfr_ethdev)
+{
+       int rc;
+       struct bnxt_vf_representor *vfr = vfr_ethdev->data->dev_private;
+       struct rte_eth_dev *parent_dev = vfr->parent_dev;
+       struct bnxt *parent_bp = parent_dev->data->dev_private;
+       uint16_t vfr_port_id = vfr_ethdev->data->port_id;
+       struct ulp_tlv_param param_list[] = {
+               {
+                       .type = BNXT_ULP_DF_PARAM_TYPE_DEV_PORT_ID,
+                       .length = 2,
+                       .value = {(vfr_port_id >> 8) & 0xff, vfr_port_id & 0xff}
+               },
+               {
+                       .type = BNXT_ULP_DF_PARAM_TYPE_LAST,
+                       .length = 0,
+                       .value = {0}
+               }
+       };
+
+       ulp_port_db_dev_port_intf_update(parent_bp->ulp_ctx, vfr_ethdev);
+
+       rc = ulp_default_flow_create(parent_dev, param_list,
+                                    BNXT_ULP_DF_TPL_VFREP_TO_VF,
+                                    &vfr->rep2vf_flow_id);
+       if (rc) {
+               BNXT_TF_DBG(DEBUG,
+                           "Default flow rule creation for VFR->VF failed!\n");
+               return -EIO;
+       }
+
+       BNXT_TF_DBG(DEBUG, "*** Default flow rule created for VFR->VF! ***\n");
+       BNXT_TF_DBG(DEBUG, "rep2vf_flow_id = %d\n", vfr->rep2vf_flow_id);
+       rc = ulp_default_flow_db_cfa_action_get(parent_bp->ulp_ctx,
+                                               vfr->rep2vf_flow_id,
+                                               &vfr->vfr_tx_cfa_action);
+       if (rc) {
+               BNXT_TF_DBG(DEBUG,
+                           "Failed to get action_ptr for VFR->VF dflt rule\n");
+               return -EIO;
+       }
+       BNXT_TF_DBG(DEBUG, "tx_cfa_action = %d\n", vfr->vfr_tx_cfa_action);
+       rc = ulp_default_flow_create(parent_dev, param_list,
+                                    BNXT_ULP_DF_TPL_VF_TO_VFREP,
+                                    &vfr->vf2rep_flow_id);
+       if (rc) {
+               BNXT_TF_DBG(DEBUG,
+                           "Default flow rule creation for VF->VFR failed!\n");
+               return -EIO;
+       }
+
+       BNXT_TF_DBG(DEBUG, "*** Default flow rule created for VF->VFR! ***\n");
+       BNXT_TF_DBG(DEBUG, "vfr2rep_flow_id = %d\n", vfr->vf2rep_flow_id);
+
+       return 0;
+}
+
+static int bnxt_vfr_alloc(struct rte_eth_dev *vfr_ethdev)
 {
        int rc = 0;
-       struct bnxt *parent_bp;
+       struct bnxt_vf_representor *vfr = vfr_ethdev->data->dev_private;
 
        if (!vfr || !vfr->parent_dev) {
                PMD_DRV_LOG(ERR,
@@ -263,10 +313,8 @@ static int bnxt_vfr_alloc(struct bnxt_vf_representor *vfr)
                return -ENOMEM;
        }
 
-       parent_bp = vfr->parent_dev->data->dev_private;
-
        /* Check if representor has been already allocated in FW */
-       if (vfr->tx_cfa_action && vfr->rx_cfa_code)
+       if (vfr->vfr_tx_cfa_action && vfr->rx_cfa_code)
                return 0;
 
        /*
@@ -274,24 +322,14 @@ static int bnxt_vfr_alloc(struct bnxt_vf_representor *vfr)
         * Otherwise the FW will create the VF-rep rules with
         * default drop action.
         */
-
-       /*
-        * This is where we need to replace invoking an HWRM cmd
-        * with the new TFLIB ULP API to do more/less the same job
-       rc = bnxt_hwrm_cfa_vfr_alloc(parent_bp,
-                                    vfr->vf_id,
-                                    &vfr->tx_cfa_action,
-                                    &vfr->rx_cfa_code);
-        */
-       if (!rc) {
-               parent_bp->cfa_code_map[vfr->rx_cfa_code] = vfr->vf_id;
+       rc = bnxt_tf_vfr_alloc(vfr_ethdev);
+       if (!rc)
                PMD_DRV_LOG(DEBUG, "allocated representor %d in FW\n",
                            vfr->vf_id);
-       } else {
+       else
                PMD_DRV_LOG(ERR,
                            "Failed to alloc representor %d in FW\n",
                            vfr->vf_id);
-       }
 
        return rc;
 }
@@ -312,7 +350,7 @@ int bnxt_vf_rep_dev_start_op(struct rte_eth_dev *eth_dev)
        struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private;
        int rc;
 
-       rc = bnxt_vfr_alloc(rep_bp);
+       rc = bnxt_vfr_alloc(eth_dev);
 
        if (!rc) {
                eth_dev->rx_pkt_burst = &bnxt_vf_rep_rx_burst;
@@ -327,6 +365,25 @@ int bnxt_vf_rep_dev_start_op(struct rte_eth_dev *eth_dev)
        return rc;
 }
 
+static int bnxt_tf_vfr_free(struct bnxt_vf_representor *vfr)
+{
+       int rc = 0;
+
+       rc = ulp_default_flow_destroy(vfr->parent_dev,
+                                     vfr->rep2vf_flow_id);
+       if (rc)
+               PMD_DRV_LOG(ERR,
+                           "default flow destroy failed rep2vf flowid: %d\n",
+                           vfr->rep2vf_flow_id);
+       rc = ulp_default_flow_destroy(vfr->parent_dev,
+                                     vfr->vf2rep_flow_id);
+       if (rc)
+               PMD_DRV_LOG(ERR,
+                           "default flow destroy failed vf2rep flowid: %d\n",
+                           vfr->vf2rep_flow_id);
+       return 0;
+}
+
 static int bnxt_vfr_free(struct bnxt_vf_representor *vfr)
 {
        int rc = 0;
@@ -341,15 +398,10 @@ static int bnxt_vfr_free(struct bnxt_vf_representor *vfr)
        parent_bp = vfr->parent_dev->data->dev_private;
 
        /* Check if representor has been already freed in FW */
-       if (!vfr->tx_cfa_action && !vfr->rx_cfa_code)
+       if (!vfr->vfr_tx_cfa_action && !vfr->rx_cfa_code)
                return 0;
 
-       /*
-        * This is where we need to replace invoking an HWRM cmd
-        * with the new TFLIB ULP API to do more/less the same job
-       rc = bnxt_hwrm_cfa_vfr_free(parent_bp,
-                                   vfr->vf_id);
-        */
+       rc = bnxt_tf_vfr_free(vfr);
        if (rc) {
                PMD_DRV_LOG(ERR,
                            "Failed to free representor %d in FW\n",
@@ -360,7 +412,7 @@ static int bnxt_vfr_free(struct bnxt_vf_representor *vfr)
        parent_bp->cfa_code_map[vfr->rx_cfa_code] = BNXT_VF_IDX_INVALID;
        PMD_DRV_LOG(DEBUG, "freed representor %d in FW\n",
                    vfr->vf_id);
-       vfr->tx_cfa_action = 0;
+       vfr->vfr_tx_cfa_action = 0;
        vfr->rx_cfa_code = 0;
 
        return rc;
index 5c2e0a0..418b95a 100644 (file)
@@ -13,8 +13,7 @@
 #define BNXT_VF_IDX_INVALID             0xffff
 
 uint16_t
-bnxt_vfr_recv(struct bnxt *bp, uint16_t cfa_code, uint16_t queue_id,
-             struct rte_mbuf *mbuf);
+bnxt_vfr_recv(uint16_t port_id, uint16_t queue_id, struct rte_mbuf *mbuf);
 int bnxt_vf_representor_init(struct rte_eth_dev *eth_dev, void *params);
 int bnxt_vf_representor_uninit(struct rte_eth_dev *eth_dev);
 int bnxt_vf_rep_dev_info_get_op(struct rte_eth_dev *eth_dev,
index 37b534f..6405887 100644 (file)
@@ -403,9 +403,9 @@ bnxt_get_rx_ts_thor(struct bnxt *bp, uint32_t rx_ts_cmpl)
 }
 #endif
 
-static void
+static uint32_t
 bnxt_ulp_set_mark_in_mbuf(struct bnxt *bp, struct rx_pkt_cmpl_hi *rxcmp1,
-                         struct rte_mbuf *mbuf)
+                         struct rte_mbuf *mbuf, uint32_t *vfr_flag)
 {
        uint32_t cfa_code;
        uint32_t meta_fmt;
@@ -415,8 +415,6 @@ bnxt_ulp_set_mark_in_mbuf(struct bnxt *bp, struct rx_pkt_cmpl_hi *rxcmp1,
        uint32_t flags2;
        uint32_t gfid_support = 0;
        int rc;
-       uint32_t vfr_flag;
-
 
        if (BNXT_GFID_ENABLED(bp))
                gfid_support = 1;
@@ -485,19 +483,21 @@ bnxt_ulp_set_mark_in_mbuf(struct bnxt *bp, struct rx_pkt_cmpl_hi *rxcmp1,
        }
 
        rc = ulp_mark_db_mark_get(bp->ulp_ctx, gfid,
-                                 cfa_code, &vfr_flag, &mark_id);
+                                 cfa_code, vfr_flag, &mark_id);
        if (!rc) {
                /* Got the mark, write it to the mbuf and return */
                mbuf->hash.fdir.hi = mark_id;
                mbuf->udata64 = (cfa_code & 0xffffffffull) << 32;
                mbuf->hash.fdir.id = rxcmp1->cfa_code;
                mbuf->ol_flags |= PKT_RX_FDIR | PKT_RX_FDIR_ID;
-               return;
+               return mark_id;
        }
 
 skip_mark:
        mbuf->hash.fdir.hi = 0;
        mbuf->hash.fdir.id = 0;
+
+       return 0;
 }
 
 void bnxt_set_mark_in_mbuf(struct bnxt *bp,
@@ -553,7 +553,7 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
        int rc = 0;
        uint8_t agg_buf = 0;
        uint16_t cmp_type;
-       uint32_t flags2_f = 0;
+       uint32_t flags2_f = 0, vfr_flag = 0, mark_id = 0;
        uint16_t flags_type;
        struct bnxt *bp = rxq->bp;
 
@@ -632,7 +632,8 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
        }
 
        if (BNXT_TRUFLOW_EN(bp))
-               bnxt_ulp_set_mark_in_mbuf(rxq->bp, rxcmp1, mbuf);
+               mark_id = bnxt_ulp_set_mark_in_mbuf(rxq->bp, rxcmp1, mbuf,
+                                                   &vfr_flag);
        else
                bnxt_set_mark_in_mbuf(rxq->bp, rxcmp1, mbuf);
 
@@ -736,10 +737,10 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt,
 rx:
        *rx_pkt = mbuf;
 
-       if ((BNXT_VF_IS_TRUSTED(rxq->bp) || BNXT_PF(rxq->bp)) &&
-           rxq->bp->cfa_code_map && rxcmp1->cfa_code) {
-               if (!bnxt_vfr_recv(rxq->bp, rxcmp1->cfa_code, rxq->queue_id,
-                                  mbuf)) {
+       if (BNXT_TRUFLOW_EN(bp) &&
+           (BNXT_VF_IS_TRUSTED(bp) || BNXT_PF(bp)) &&
+           vfr_flag) {
+               if (!bnxt_vfr_recv(mark_id, rxq->queue_id, mbuf)) {
                        /* Now return an error so that nb_rx_pkts is not
                         * incremented.
                         * This packet was meant to be given to the representor.
index 69ff89a..a1ab3f3 100644 (file)
@@ -30,6 +30,7 @@ struct bnxt_tx_queue {
        int                     index;
        int                     tx_wake_thresh;
        uint32_t                tx_cfa_action;
+       uint32_t                vfr_tx_cfa_action;
        struct bnxt_tx_ring_info        *tx_ring;
 
        unsigned int            cp_nr_rings;