net/mlx5: fix RSS expansion for GTP
[dpdk.git] / drivers / net / ixgbe / ixgbe_pf.c
index c88d56e..fbf2b17 100644 (file)
@@ -15,7 +15,7 @@
 #include <rte_debug.h>
 #include <rte_eal.h>
 #include <rte_ether.h>
-#include <rte_ethdev_driver.h>
+#include <ethdev_driver.h>
 #include <rte_memcpy.h>
 #include <rte_malloc.h>
 #include <rte_random.h>
@@ -66,7 +66,7 @@ ixgbe_mb_intr_setup(struct rte_eth_dev *dev)
        return 0;
 }
 
-void ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
+int ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
 {
        struct ixgbe_vf_info **vfinfo =
                IXGBE_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
@@ -78,19 +78,30 @@ void ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
                IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
        uint16_t vf_num;
        uint8_t nb_queue;
+       int ret = 0;
 
        PMD_INIT_FUNC_TRACE();
 
        RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
        vf_num = dev_num_vf(eth_dev);
        if (vf_num == 0)
-               return;
+               return ret;
 
        *vfinfo = rte_zmalloc("vf_info", sizeof(struct ixgbe_vf_info) * vf_num, 0);
-       if (*vfinfo == NULL)
-               rte_panic("Cannot allocate memory for private VF data\n");
+       if (*vfinfo == NULL) {
+               PMD_INIT_LOG(ERR,
+                       "Cannot allocate memory for private VF data");
+               return -ENOMEM;
+       }
 
-       rte_eth_switch_domain_alloc(&(*vfinfo)->switch_domain_id);
+       ret = rte_eth_switch_domain_alloc(&(*vfinfo)->switch_domain_id);
+       if (ret) {
+               PMD_INIT_LOG(ERR,
+                       "failed to allocate switch domain for device %d", ret);
+               rte_free(*vfinfo);
+               *vfinfo = NULL;
+               return ret;
+       }
 
        memset(mirror_info, 0, sizeof(struct ixgbe_mirror_info));
        memset(uta_info, 0, sizeof(struct ixgbe_uta_info));
@@ -118,6 +129,8 @@ void ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
 
        /* set mb interrupt mask */
        ixgbe_mb_intr_setup(eth_dev);
+
+       return ret;
 }
 
 void ixgbe_pf_host_uninit(struct rte_eth_dev *eth_dev)
@@ -161,16 +174,14 @@ ixgbe_add_tx_flow_control_drop_filter(struct rte_eth_dev *eth_dev)
        struct ixgbe_ethertype_filter ethertype_filter;
 
        if (!hw->mac.ops.set_ethertype_anti_spoofing) {
-               RTE_LOG(INFO, PMD, "ether type anti-spoofing is not"
-                       " supported.\n");
+               PMD_DRV_LOG(INFO, "ether type anti-spoofing is not supported.\n");
                return;
        }
 
        i = ixgbe_ethertype_filter_lookup(filter_info,
                                          IXGBE_ETHERTYPE_FLOW_CTRL);
        if (i >= 0) {
-               RTE_LOG(ERR, PMD, "A ether type filter"
-                       " entity for flow control already exists!\n");
+               PMD_DRV_LOG(ERR, "A ether type filter entity for flow control already exists!\n");
                return;
        }
 
@@ -183,8 +194,7 @@ ixgbe_add_tx_flow_control_drop_filter(struct rte_eth_dev *eth_dev)
        i = ixgbe_ethertype_filter_insert(filter_info,
                                          &ethertype_filter);
        if (i < 0) {
-               RTE_LOG(ERR, PMD, "Cannot find an unused ether type filter"
-                       " entity for flow control.\n");
+               PMD_DRV_LOG(ERR, "Cannot find an unused ether type filter entity for flow control.\n");
                return;
        }
 
@@ -415,7 +425,7 @@ ixgbe_disable_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf)
 
        vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
 
-       RTE_LOG(INFO, PMD, "VF %u: disabling multicast promiscuous\n", vf);
+       PMD_DRV_LOG(INFO, "VF %u: disabling multicast promiscuous\n", vf);
 
        vmolr &= ~IXGBE_VMOLR_MPE;
 
@@ -542,19 +552,47 @@ ixgbe_vf_set_vlan(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 }
 
 static int
-ixgbe_set_vf_lpe(struct rte_eth_dev *dev, __rte_unused uint32_t vf, uint32_t *msgbuf)
+ixgbe_set_vf_lpe(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
 {
        struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-       uint32_t new_mtu = msgbuf[1];
+       uint32_t max_frame = msgbuf[1];
        uint32_t max_frs;
-       int max_frame = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
+       uint32_t hlreg0;
 
        /* X540 and X550 support jumbo frames in IOV mode */
        if (hw->mac.type != ixgbe_mac_X540 &&
                hw->mac.type != ixgbe_mac_X550 &&
                hw->mac.type != ixgbe_mac_X550EM_x &&
-               hw->mac.type != ixgbe_mac_X550EM_a)
-               return -1;
+               hw->mac.type != ixgbe_mac_X550EM_a) {
+               struct ixgbe_vf_info *vfinfo =
+                       *IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
+
+               switch (vfinfo[vf].api_version) {
+               case ixgbe_mbox_api_11:
+               case ixgbe_mbox_api_12:
+               case ixgbe_mbox_api_13:
+                        /**
+                         * Version 1.1&1.2&1.3 supports jumbo frames on VFs
+                         * if PF has jumbo frames enabled which means legacy
+                         * VFs are disabled.
+                         */
+                       if (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+                           IXGBE_ETH_MAX_LEN)
+                               break;
+                       /* fall through */
+               default:
+                       /**
+                        * If the PF or VF are running w/ jumbo frames enabled,
+                        * we return -1 as we cannot support jumbo frames on
+                        * legacy VFs.
+                        */
+                       if (max_frame > IXGBE_ETH_MAX_LEN ||
+                           dev->data->dev_conf.rxmode.max_rx_pkt_len >
+                           IXGBE_ETH_MAX_LEN)
+                               return -1;
+                       break;
+               }
+       }
 
        if (max_frame < RTE_ETHER_MIN_LEN ||
                        max_frame > RTE_ETHER_MAX_JUMBO_FRAME_LEN)
@@ -562,8 +600,20 @@ ixgbe_set_vf_lpe(struct rte_eth_dev *dev, __rte_unused uint32_t vf, uint32_t *ms
 
        max_frs = (IXGBE_READ_REG(hw, IXGBE_MAXFRS) &
                   IXGBE_MHADD_MFS_MASK) >> IXGBE_MHADD_MFS_SHIFT;
-       if (max_frs < new_mtu) {
-               max_frs = new_mtu << IXGBE_MHADD_MFS_SHIFT;
+       if (max_frs < max_frame) {
+               hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
+               if (max_frame > IXGBE_ETH_MAX_LEN) {
+                       dev->data->dev_conf.rxmode.offloads |=
+                               DEV_RX_OFFLOAD_JUMBO_FRAME;
+                       hlreg0 |= IXGBE_HLREG0_JUMBOEN;
+               } else {
+                       dev->data->dev_conf.rxmode.offloads &=
+                               ~DEV_RX_OFFLOAD_JUMBO_FRAME;
+                       hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
+               }
+               IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
+
+               max_frs = max_frame << IXGBE_MHADD_MFS_SHIFT;
                IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, max_frs);
        }
 
@@ -588,7 +638,7 @@ ixgbe_negotiate_vf_api(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
                break;
        }
 
-       RTE_LOG(ERR, PMD, "Negotiate invalid api version %u from VF %d\n",
+       PMD_DRV_LOG(ERR, "Negotiate invalid api version %u from VF %d\n",
                api_version, vf);
 
        return -1;
@@ -618,6 +668,7 @@ ixgbe_get_vf_queues(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
        case ixgbe_mbox_api_20:
        case ixgbe_mbox_api_11:
        case ixgbe_mbox_api_12:
+       case ixgbe_mbox_api_13:
                break;
        default:
                return -1;
@@ -635,7 +686,7 @@ ixgbe_get_vf_queues(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
        switch (eth_conf->txmode.mq_mode) {
        case ETH_MQ_TX_NONE:
        case ETH_MQ_TX_DCB:
-               RTE_LOG(ERR, PMD, "PF must work with virtualization for VF %u"
+               PMD_DRV_LOG(ERR, "PF must work with virtualization for VF %u"
                        ", but its tx mode = %d\n", vf,
                        eth_conf->txmode.mq_mode);
                return -1;
@@ -670,7 +721,7 @@ ixgbe_get_vf_queues(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
                break;
 
        default:
-               RTE_LOG(ERR, PMD, "PF work with invalid mode = %d\n",
+               PMD_DRV_LOG(ERR, "PF work with invalid mode = %d\n",
                        eth_conf->txmode.mq_mode);
                return -1;
        }
@@ -725,7 +776,7 @@ ixgbe_set_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
                fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
                if (!(fctrl & IXGBE_FCTRL_UPE)) {
                        /* VF promisc requires PF in promisc */
-                       RTE_LOG(ERR, PMD,
+                       PMD_DRV_LOG(ERR,
                               "Enabling VF promisc requires PF in promisc\n");
                        return -1;
                }
@@ -750,6 +801,36 @@ out:
        return 0;
 }
 
+static int
+ixgbe_set_vf_macvlan_msg(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
+{
+       struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct ixgbe_vf_info *vf_info =
+               *(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+       uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
+       int index = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK) >>
+                   IXGBE_VT_MSGINFO_SHIFT;
+
+       if (index) {
+               if (!rte_is_valid_assigned_ether_addr(
+                       (struct rte_ether_addr *)new_mac)) {
+                       PMD_DRV_LOG(ERR, "set invalid mac vf:%d\n", vf);
+                       return -1;
+               }
+
+               vf_info[vf].mac_count++;
+
+               hw->mac.ops.set_rar(hw, vf_info[vf].mac_count,
+                               new_mac, vf, IXGBE_RAH_AV);
+       } else {
+               if (vf_info[vf].mac_count) {
+                       hw->mac.ops.clear_rar(hw, vf_info[vf].mac_count);
+                       vf_info[vf].mac_count = 0;
+               }
+       }
+       return 0;
+}
+
 static int
 ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
 {
@@ -791,7 +872,7 @@ ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
                vfinfo[vf].clear_to_send = true;
 
                /* notify application about VF reset */
-               _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX,
+               rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX,
                                              &ret_param);
                return ret;
        }
@@ -803,8 +884,7 @@ ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
         * if 0, do nothing and send ACK to VF
         * if ret_param.retval > 1, do nothing and send NAK to VF
         */
-       _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX,
-                                     &ret_param);
+       rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX, &ret_param);
 
        retval = ret_param.retval;
 
@@ -837,6 +917,10 @@ ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
                if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
                        retval = ixgbe_set_vf_mc_promisc(dev, vf, msgbuf);
                break;
+       case IXGBE_VF_SET_MACVLAN:
+               if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
+                       retval = ixgbe_set_vf_macvlan_msg(dev, vf, msgbuf);
+               break;
        default:
                PMD_DRV_LOG(DEBUG, "Unhandled Msg %8.8x", (unsigned)msgbuf[0]);
                retval = IXGBE_ERR_MBX;