How to implement the "setting mac address " for VF which is a qdma device using QDMA-DPDK testpmd application

[vijayaKesavan19]

Hi team,

I am trying to do port representor testing on QDMA-DPDK using the openNIC design project. github link for the openNIC design project

For that I need to test changing the mac address of the PF and VF (from the PF).

For setting MAC of PF I use this command testpmd> mac_addr set <port id> <XX:XX:XX:XX:XX:XX>

I was able to do that by writing a function which is given below

int qdma_set_mac_address(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
{
        int i ;
        uint64_t mac_addr = (uint64_t)(uintptr_t)&addr;
        if (!rte_is_valid_assigned_ether_addr(addr))
               return -EADDRNOTAVAIL;
        for (i = 0; i < RTE_ETHER_ADDR_LEN; i++)
                dev->data->mac_addrs[0].addr_bytes[RTE_ETHER_ADDR_LEN -1 -i] =
                                                (uint8_t)((mac_addr >> (BITS_IN_BYTE * i)) & 0xFF);
        return 0;
}

File : drivers/net/qdma/qdma_devops.c

For setting MAC of VF I use this command testpmd> set vf mac addr <port id> <vf id> <XX:XX:XX:XX:XX:XX> I hope the below funciton is called.

static void  (
        void *parsed_result,
        __rte_unused struct cmdline *cl,
        __rte_unused void *data)
{
        struct cmd_set_vf_mac_addr_result *res = parsed_result;
        int ret = -ENOTSUP;

        if (port_id_is_invalid(res->port_id, ENABLED_WARN))
                return;

#ifdef RTE_NET_IXGBE
        if (ret == -ENOTSUP)
                ret = rte_pmd_ixgbe_set_vf_mac_addr(res->port_id, res->vf_id,
                                &res->mac_addr);
#endif
#ifdef RTE_NET_I40E
        if (ret == -ENOTSUP)
                ret = rte_pmd_i40e_set_vf_mac_addr(res->port_id, res->vf_id,
                                &res->mac_addr);
#endif
#ifdef RTE_NET_BNXT
        if (ret == -ENOTSUP)
                ret = rte_pmd_bnxt_set_vf_mac_addr(res->port_id, res->vf_id,
                                &res->mac_addr);
#endif
     
        switch (ret) {
        case 0:
                break;
        case -EINVAL:
                printf("invalid vf_id %d or mac_addr\n", res->vf_id);
                break;
        case -ENODEV:
                printf("invalid port_id %d\n", res->port_id);
                break;
        case -ENOTSUP:
                printf("function not implemented\n");
                break;
        default:
                printf("programming error: (%s)\n", strerror(-ret));
        }
}

File : dpdk-20.11/app/test-pmd/cmdline.c Line: 11427

Which is supported only when the pf and vf is ixgbe, i40e or bnxt device. But what I am using is qdma device (i.e: device name of the VF is "qdma_vf" ).

The below is the one of the driver function for setting mac address

int
rte_pmd_ixgbe_set_vf_mac_addr(uint16_t port, uint16_t vf,
                              struct rte_ether_addr *mac_addr)
{
        struct ixgbe_hw *hw;
        struct ixgbe_vf_info *vfinfo;
        int rar_entry;
        uint8_t *new_mac = (uint8_t *)(mac_addr);
        struct rte_eth_dev *dev;
        struct rte_pci_device *pci_dev;

        RTE_ETH_VALID_PORTID_OR_ERR_RET(port, -ENODEV);

        dev = &rte_eth_devices[port];
        pci_dev = RTE_ETH_DEV_TO_PCI(dev);

        if (!is_ixgbe_supported(dev))
                return -ENOTSUP;

        if (vf >= pci_dev->max_vfs)
                return -EINVAL;

        **hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
        vfinfo = *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));**
        rar_entry = hw->mac.num_rar_entries - (vf + 1);

        if (rte_is_valid_assigned_ether_addr(
                        (struct rte_ether_addr *)new_mac)) {
                rte_memcpy(vfinfo[vf].vf_mac_addresses, new_mac,
                           RTE_ETHER_ADDR_LEN);
                return hw->mac.ops.set_rar(hw, rar_entry, new_mac, vf,
                                           IXGBE_RAH_AV);
        }
        return -EINVAL;
}

File : drivers/net/ixgbe/rte_pmd_ixgbe.c Line: 1174

As you can see the dev->data->private member was casted by the ixgbe driver preference to fetch the VF's info. Structure of ixgbe_vf_info

struct ixgbe_vf_info {
        uint8_t vf_mac_addresses[RTE_ETHER_ADDR_LEN];
        uint16_t vf_mc_hashes[IXGBE_MAX_VF_MC_ENTRIES];
        uint16_t num_vf_mc_hashes;
        uint16_t default_vf_vlan_id;
        uint16_t vlans_enabled;
        bool clear_to_send;
        uint16_t tx_rate[IXGBE_MAX_QUEUE_NUM_PER_VF];
        uint16_t vlan_count;
        uint8_t spoofchk_enabled;
        uint8_t api_version;
        uint16_t switch_domain_id;
        uint16_t xcast_mode;
        uint16_t mac_count;
};

In qdma driver I can only fetch the function id of the VF on the vfinfo.

struct qdma_vf_info {
        uint16_t        func_id;
};

Can you tell me how to access the mac address and received packets of the qdma_vf device(VF) through the physical function(PF) on QDMA-DPDK testpmd application.

This issue is causing a major disruption in our production environment. Could you please consider labeling it as high priority?

Thank you in advance