net/igc: implement device base operations
authorAlvin Zhang <alvinx.zhang@intel.com>
Wed, 15 Apr 2020 08:48:02 +0000 (16:48 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 21 Apr 2020 11:57:07 +0000 (13:57 +0200)
Bellow ops are implemented:
dev_configure
dev_start
dev_stop
dev_close
dev_reset
dev_set_link_up
dev_set_link_down
link_update
fw_version_get
dev_led_on
dev_led_off

Signed-off-by: Alvin Zhang <alvinx.zhang@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
doc/guides/nics/features/igc.ini
drivers/net/igc/igc_ethdev.c
drivers/net/igc/igc_ethdev.h

index ad75cc4..0fbdf7c 100644 (file)
@@ -3,6 +3,11 @@
 ; Refer to default.ini for the full list of available PMD features.
 ;
 [Features]
+Speed capabilities   = Y
+Link status          = Y
+Link status event    = Y
+FW version           = Y
+LED                  = Y
 Linux UIO            = Y
 Linux VFIO           = Y
 x86-64               = Y
index f23543e..423c817 100644 (file)
@@ -3,6 +3,7 @@
  */
 
 #include <stdint.h>
+#include <string.h>
 
 #include <rte_pci.h>
 #include <rte_bus_pci.h>
 
 #define IGC_INTEL_VENDOR_ID            0x8086
 
+/*
+ * The overhead from MTU to max frame size.
+ * Considering VLAN so tag needs to be counted.
+ */
+#define IGC_ETH_OVERHEAD               (RTE_ETHER_HDR_LEN + \
+                                       RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE)
+
 #define IGC_FC_PAUSE_TIME              0x0680
+#define IGC_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
+#define IGC_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
+
+#define IGC_MISC_VEC_ID                        RTE_INTR_VEC_ZERO_OFFSET
+#define IGC_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
+#define IGC_MSIX_OTHER_INTR_VEC                0   /* MSI-X other interrupt vector */
+#define IGC_FLAG_NEED_LINK_UPDATE      (1u << 0)       /* need update link */
+
+#define IGC_DEFAULT_RX_FREE_THRESH     32
+
+#define IGC_DEFAULT_RX_PTHRESH         8
+#define IGC_DEFAULT_RX_HTHRESH         8
+#define IGC_DEFAULT_RX_WTHRESH         4
+
+#define IGC_DEFAULT_TX_PTHRESH         8
+#define IGC_DEFAULT_TX_HTHRESH         1
+#define IGC_DEFAULT_TX_WTHRESH         16
+
+/* MSI-X other interrupt vector */
+#define IGC_MSIX_OTHER_INTR_VEC                0
 
 static const struct rte_pci_id pci_id_igc_map[] = {
        { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_LM) },
@@ -29,12 +57,20 @@ static int eth_igc_configure(struct rte_eth_dev *dev);
 static int eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 static void eth_igc_stop(struct rte_eth_dev *dev);
 static int eth_igc_start(struct rte_eth_dev *dev);
+static int eth_igc_set_link_up(struct rte_eth_dev *dev);
+static int eth_igc_set_link_down(struct rte_eth_dev *dev);
 static void eth_igc_close(struct rte_eth_dev *dev);
 static int eth_igc_reset(struct rte_eth_dev *dev);
 static int eth_igc_promiscuous_enable(struct rte_eth_dev *dev);
 static int eth_igc_promiscuous_disable(struct rte_eth_dev *dev);
+static int eth_igc_fw_version_get(struct rte_eth_dev *dev,
+                               char *fw_version, size_t fw_size);
 static int eth_igc_infos_get(struct rte_eth_dev *dev,
                        struct rte_eth_dev_info *dev_info);
+static int eth_igc_led_on(struct rte_eth_dev *dev);
+static int eth_igc_led_off(struct rte_eth_dev *dev);
+static void eth_igc_tx_queue_release(void *txq);
+static void eth_igc_rx_queue_release(void *rxq);
 static int
 eth_igc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
                uint16_t nb_rx_desc, unsigned int socket_id,
@@ -52,35 +88,394 @@ static const struct eth_dev_ops eth_igc_ops = {
        .dev_start              = eth_igc_start,
        .dev_close              = eth_igc_close,
        .dev_reset              = eth_igc_reset,
+       .dev_set_link_up        = eth_igc_set_link_up,
+       .dev_set_link_down      = eth_igc_set_link_down,
        .promiscuous_enable     = eth_igc_promiscuous_enable,
        .promiscuous_disable    = eth_igc_promiscuous_disable,
+
+       .fw_version_get         = eth_igc_fw_version_get,
        .dev_infos_get          = eth_igc_infos_get,
+       .dev_led_on             = eth_igc_led_on,
+       .dev_led_off            = eth_igc_led_off,
+
        .rx_queue_setup         = eth_igc_rx_queue_setup,
+       .rx_queue_release       = eth_igc_rx_queue_release,
        .tx_queue_setup         = eth_igc_tx_queue_setup,
+       .tx_queue_release       = eth_igc_tx_queue_release,
 };
 
+/*
+ * multiple queue mode checking
+ */
+static int
+igc_check_mq_mode(struct rte_eth_dev *dev)
+{
+       enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
+       enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
+
+       if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
+               PMD_INIT_LOG(ERR, "SRIOV is not supported.");
+               return -EINVAL;
+       }
+
+       if (rx_mq_mode != ETH_MQ_RX_NONE &&
+               rx_mq_mode != ETH_MQ_RX_RSS) {
+               /* RSS together with VMDq not supported*/
+               PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
+                               rx_mq_mode);
+               return -EINVAL;
+       }
+
+       /* To no break software that set invalid mode, only display
+        * warning if invalid mode is used.
+        */
+       if (tx_mq_mode != ETH_MQ_TX_NONE)
+               PMD_INIT_LOG(WARNING,
+                       "TX mode %d is not supported. Due to meaningless in this driver, just ignore",
+                       tx_mq_mode);
+
+       return 0;
+}
+
 static int
 eth_igc_configure(struct rte_eth_dev *dev)
 {
+       struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
+       int ret;
+
        PMD_INIT_FUNC_TRACE();
-       RTE_SET_USED(dev);
+
+       ret  = igc_check_mq_mode(dev);
+       if (ret != 0)
+               return ret;
+
+       intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
        return 0;
 }
 
 static int
-eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete)
+eth_igc_set_link_up(struct rte_eth_dev *dev)
 {
-       PMD_INIT_FUNC_TRACE();
-       RTE_SET_USED(dev);
-       RTE_SET_USED(wait_to_complete);
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       if (hw->phy.media_type == igc_media_type_copper)
+               igc_power_up_phy(hw);
+       else
+               igc_power_up_fiber_serdes_link(hw);
+       return 0;
+}
+
+static int
+eth_igc_set_link_down(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       if (hw->phy.media_type == igc_media_type_copper)
+               igc_power_down_phy(hw);
+       else
+               igc_shutdown_fiber_serdes_link(hw);
        return 0;
 }
 
+/*
+ * disable other interrupt
+ */
+static void
+igc_intr_other_disable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+
+       if (rte_intr_allow_others(intr_handle) &&
+               dev->data->dev_conf.intr_conf.lsc) {
+               IGC_WRITE_REG(hw, IGC_EIMC, 1u << IGC_MSIX_OTHER_INTR_VEC);
+       }
+
+       IGC_WRITE_REG(hw, IGC_IMC, ~0);
+       IGC_WRITE_FLUSH(hw);
+}
+
+/*
+ * enable other interrupt
+ */
+static inline void
+igc_intr_other_enable(struct rte_eth_dev *dev)
+{
+       struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+
+       if (rte_intr_allow_others(intr_handle) &&
+               dev->data->dev_conf.intr_conf.lsc) {
+               IGC_WRITE_REG(hw, IGC_EIMS, 1u << IGC_MSIX_OTHER_INTR_VEC);
+       }
+
+       IGC_WRITE_REG(hw, IGC_IMS, intr->mask);
+       IGC_WRITE_FLUSH(hw);
+}
+
+/*
+ * It reads ICR and gets interrupt causes, check it and set a bit flag
+ * to update link status.
+ */
+static void
+eth_igc_interrupt_get_status(struct rte_eth_dev *dev)
+{
+       uint32_t icr;
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
+
+       /* read-on-clear nic registers here */
+       icr = IGC_READ_REG(hw, IGC_ICR);
+
+       intr->flags = 0;
+       if (icr & IGC_ICR_LSC)
+               intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
+}
+
+/* return 0 means link status changed, -1 means not changed */
+static int
+eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_eth_link link;
+       int link_check, count;
+
+       link_check = 0;
+       hw->mac.get_link_status = 1;
+
+       /* possible wait-to-complete in up to 9 seconds */
+       for (count = 0; count < IGC_LINK_UPDATE_CHECK_TIMEOUT; count++) {
+               /* Read the real link status */
+               switch (hw->phy.media_type) {
+               case igc_media_type_copper:
+                       /* Do the work to read phy */
+                       igc_check_for_link(hw);
+                       link_check = !hw->mac.get_link_status;
+                       break;
+
+               case igc_media_type_fiber:
+                       igc_check_for_link(hw);
+                       link_check = (IGC_READ_REG(hw, IGC_STATUS) &
+                                     IGC_STATUS_LU);
+                       break;
+
+               case igc_media_type_internal_serdes:
+                       igc_check_for_link(hw);
+                       link_check = hw->mac.serdes_has_link;
+                       break;
+
+               default:
+                       break;
+               }
+               if (link_check || wait_to_complete == 0)
+                       break;
+               rte_delay_ms(IGC_LINK_UPDATE_CHECK_INTERVAL);
+       }
+       memset(&link, 0, sizeof(link));
+
+       /* Now we check if a transition has happened */
+       if (link_check) {
+               uint16_t duplex, speed;
+               hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
+               link.link_duplex = (duplex == FULL_DUPLEX) ?
+                               ETH_LINK_FULL_DUPLEX :
+                               ETH_LINK_HALF_DUPLEX;
+               link.link_speed = speed;
+               link.link_status = ETH_LINK_UP;
+               link.link_autoneg = !(dev->data->dev_conf.link_speeds &
+                               ETH_LINK_SPEED_FIXED);
+
+               if (speed == SPEED_2500) {
+                       uint32_t tipg = IGC_READ_REG(hw, IGC_TIPG);
+                       if ((tipg & IGC_TIPG_IPGT_MASK) != 0x0b) {
+                               tipg &= ~IGC_TIPG_IPGT_MASK;
+                               tipg |= 0x0b;
+                               IGC_WRITE_REG(hw, IGC_TIPG, tipg);
+                       }
+               }
+       } else {
+               link.link_speed = 0;
+               link.link_duplex = ETH_LINK_HALF_DUPLEX;
+               link.link_status = ETH_LINK_DOWN;
+               link.link_autoneg = ETH_LINK_FIXED;
+       }
+
+       return rte_eth_linkstatus_set(dev, &link);
+}
+
+/*
+ * It executes link_update after knowing an interrupt is present.
+ */
+static void
+eth_igc_interrupt_action(struct rte_eth_dev *dev)
+{
+       struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_eth_link link;
+       int ret;
+
+       if (intr->flags & IGC_FLAG_NEED_LINK_UPDATE) {
+               intr->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
+
+               /* set get_link_status to check register later */
+               ret = eth_igc_link_update(dev, 0);
+
+               /* check if link has changed */
+               if (ret < 0)
+                       return;
+
+               rte_eth_linkstatus_get(dev, &link);
+               if (link.link_status)
+                       PMD_DRV_LOG(INFO,
+                               " Port %d: Link Up - speed %u Mbps - %s",
+                               dev->data->port_id,
+                               (unsigned int)link.link_speed,
+                               link.link_duplex == ETH_LINK_FULL_DUPLEX ?
+                               "full-duplex" : "half-duplex");
+               else
+                       PMD_DRV_LOG(INFO, " Port %d: Link Down",
+                               dev->data->port_id);
+
+               PMD_DRV_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
+                               pci_dev->addr.domain,
+                               pci_dev->addr.bus,
+                               pci_dev->addr.devid,
+                               pci_dev->addr.function);
+               _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC,
+                               NULL);
+       }
+}
+
+/*
+ * Interrupt handler which shall be registered at first.
+ *
+ * @handle
+ *  Pointer to interrupt handle.
+ * @param
+ *  The address of parameter (struct rte_eth_dev *) registered before.
+ */
+static void
+eth_igc_interrupt_handler(void *param)
+{
+       struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
+
+       eth_igc_interrupt_get_status(dev);
+       eth_igc_interrupt_action(dev);
+}
+
+/*
+ *  This routine disables all traffic on the adapter by issuing a
+ *  global reset on the MAC.
+ */
 static void
 eth_igc_stop(struct rte_eth_dev *dev)
 {
-       PMD_INIT_FUNC_TRACE();
-       RTE_SET_USED(dev);
+       struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       struct rte_eth_link link;
+
+       adapter->stopped = 1;
+
+       /* disable all MSI-X interrupts */
+       IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
+       IGC_WRITE_FLUSH(hw);
+
+       /* clear all MSI-X interrupts */
+       IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
+
+       igc_intr_other_disable(dev);
+
+       /* disable intr eventfd mapping */
+       rte_intr_disable(intr_handle);
+
+       igc_reset_hw(hw);
+
+       /* disable all wake up */
+       IGC_WRITE_REG(hw, IGC_WUC, 0);
+
+       /* Set bit for Go Link disconnect */
+       igc_read_reg_check_set_bits(hw, IGC_82580_PHY_POWER_MGMT,
+                       IGC_82580_PM_GO_LINKD);
+
+       /* Power down the phy. Needed to make the link go Down */
+       eth_igc_set_link_down(dev);
+
+       /* clear the recorded link status */
+       memset(&link, 0, sizeof(link));
+       rte_eth_linkstatus_set(dev, &link);
+
+       if (!rte_intr_allow_others(intr_handle))
+               /* resume to the default handler */
+               rte_intr_callback_register(intr_handle,
+                                          eth_igc_interrupt_handler,
+                                          (void *)dev);
+
+       /* Clean datapath event and queue/vec mapping */
+       rte_intr_efd_disable(intr_handle);
+}
+
+/* Sets up the hardware to generate MSI-X interrupts properly
+ * @hw
+ *  board private structure
+ */
+static void
+igc_configure_msix_intr(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+
+       uint32_t intr_mask;
+
+       /* won't configure msix register if no mapping is done
+        * between intr vector and event fd
+        */
+       if (!rte_intr_dp_is_en(intr_handle) ||
+               !dev->data->dev_conf.intr_conf.lsc)
+               return;
+
+       /* turn on MSI-X capability first */
+       IGC_WRITE_REG(hw, IGC_GPIE, IGC_GPIE_MSIX_MODE |
+                               IGC_GPIE_PBA | IGC_GPIE_EIAME |
+                               IGC_GPIE_NSICR);
+
+       intr_mask = (1u << IGC_MSIX_OTHER_INTR_VEC);
+
+       /* enable msix auto-clear */
+       igc_read_reg_check_set_bits(hw, IGC_EIAC, intr_mask);
+
+       /* set other cause interrupt vector */
+       igc_read_reg_check_set_bits(hw, IGC_IVAR_MISC,
+               (uint32_t)(IGC_MSIX_OTHER_INTR_VEC | IGC_IVAR_VALID) << 8);
+
+       /* enable auto-mask */
+       igc_read_reg_check_set_bits(hw, IGC_EIAM, intr_mask);
+
+       IGC_WRITE_FLUSH(hw);
+}
+
+/**
+ * It enables the interrupt mask and then enable the interrupt.
+ *
+ * @dev
+ *  Pointer to struct rte_eth_dev.
+ * @on
+ *  Enable or Disable
+ */
+static void
+igc_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
+{
+       struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
+
+       if (on)
+               intr->mask |= IGC_ICR_LSC;
+       else
+               intr->mask &= ~IGC_ICR_LSC;
 }
 
 /*
@@ -170,9 +565,134 @@ igc_hardware_init(struct igc_hw *hw)
 static int
 eth_igc_start(struct rte_eth_dev *dev)
 {
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       uint32_t *speeds;
+       int num_speeds;
+       bool autoneg;
+
        PMD_INIT_FUNC_TRACE();
-       RTE_SET_USED(dev);
+
+       /* disable all MSI-X interrupts */
+       IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
+       IGC_WRITE_FLUSH(hw);
+
+       /* clear all MSI-X interrupts */
+       IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
+
+       /* disable uio/vfio intr/eventfd mapping */
+       if (!adapter->stopped)
+               rte_intr_disable(intr_handle);
+
+       /* Power up the phy. Needed to make the link go Up */
+       eth_igc_set_link_up(dev);
+
+       /* Put the address into the Receive Address Array */
+       igc_rar_set(hw, hw->mac.addr, 0);
+
+       /* Initialize the hardware */
+       if (igc_hardware_init(hw)) {
+               PMD_DRV_LOG(ERR, "Unable to initialize the hardware");
+               return -EIO;
+       }
+       adapter->stopped = 0;
+
+       /* confiugre msix for rx interrupt */
+       igc_configure_msix_intr(dev);
+
+       igc_clear_hw_cntrs_base_generic(hw);
+
+       /* Setup link speed and duplex */
+       speeds = &dev->data->dev_conf.link_speeds;
+       if (*speeds == ETH_LINK_SPEED_AUTONEG) {
+               hw->phy.autoneg_advertised = IGC_ALL_SPEED_DUPLEX_2500;
+               hw->mac.autoneg = 1;
+       } else {
+               num_speeds = 0;
+               autoneg = (*speeds & ETH_LINK_SPEED_FIXED) == 0;
+
+               /* Reset */
+               hw->phy.autoneg_advertised = 0;
+
+               if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
+                               ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
+                               ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G |
+                               ETH_LINK_SPEED_FIXED)) {
+                       num_speeds = -1;
+                       goto error_invalid_config;
+               }
+               if (*speeds & ETH_LINK_SPEED_10M_HD) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
+                       num_speeds++;
+               }
+               if (*speeds & ETH_LINK_SPEED_10M) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
+                       num_speeds++;
+               }
+               if (*speeds & ETH_LINK_SPEED_100M_HD) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
+                       num_speeds++;
+               }
+               if (*speeds & ETH_LINK_SPEED_100M) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
+                       num_speeds++;
+               }
+               if (*speeds & ETH_LINK_SPEED_1G) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
+                       num_speeds++;
+               }
+               if (*speeds & ETH_LINK_SPEED_2_5G) {
+                       hw->phy.autoneg_advertised |= ADVERTISE_2500_FULL;
+                       num_speeds++;
+               }
+               if (num_speeds == 0 || (!autoneg && num_speeds > 1))
+                       goto error_invalid_config;
+
+               /* Set/reset the mac.autoneg based on the link speed,
+                * fixed or not
+                */
+               if (!autoneg) {
+                       hw->mac.autoneg = 0;
+                       hw->mac.forced_speed_duplex =
+                                       hw->phy.autoneg_advertised;
+               } else {
+                       hw->mac.autoneg = 1;
+               }
+       }
+
+       igc_setup_link(hw);
+
+       if (rte_intr_allow_others(intr_handle)) {
+               /* check if lsc interrupt is enabled */
+               if (dev->data->dev_conf.intr_conf.lsc)
+                       igc_lsc_interrupt_setup(dev, 1);
+               else
+                       igc_lsc_interrupt_setup(dev, 0);
+       } else {
+               rte_intr_callback_unregister(intr_handle,
+                                            eth_igc_interrupt_handler,
+                                            (void *)dev);
+               if (dev->data->dev_conf.intr_conf.lsc)
+                       PMD_DRV_LOG(INFO,
+                               "LSC won't enable because of no intr multiplex");
+       }
+
+       /* enable uio/vfio intr/eventfd mapping */
+       rte_intr_enable(intr_handle);
+
+       /* resume enabled intr since hw reset */
+       igc_intr_other_enable(dev);
+
+       eth_igc_link_update(dev, 0);
+
        return 0;
+
+error_invalid_config:
+       PMD_DRV_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
+                    dev->data->dev_conf.link_speeds, dev->data->port_id);
+       return -EINVAL;
 }
 
 static int
@@ -232,10 +752,28 @@ igc_reset_swfw_lock(struct igc_hw *hw)
 static void
 eth_igc_close(struct rte_eth_dev *dev)
 {
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
        struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
+       int retry = 0;
 
        PMD_INIT_FUNC_TRACE();
 
+       if (!adapter->stopped)
+               eth_igc_stop(dev);
+
+       igc_intr_other_disable(dev);
+       do {
+               int ret = rte_intr_callback_unregister(intr_handle,
+                               eth_igc_interrupt_handler, dev);
+               if (ret >= 0 || ret == -ENOENT || ret == -EINVAL)
+                       break;
+
+               PMD_DRV_LOG(ERR, "intr callback unregister failed: %d", ret);
+               DELAY(200 * 1000); /* delay 200ms */
+       } while (retry++ < 5);
+
        igc_phy_hw_reset(hw);
        igc_hw_control_release(hw);
 
@@ -258,6 +796,7 @@ static int
 eth_igc_dev_init(struct rte_eth_dev *dev)
 {
        struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
        struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
        int error = 0;
 
@@ -364,6 +903,7 @@ eth_igc_dev_init(struct rte_eth_dev *dev)
        dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 
        hw->mac.get_link_status = 1;
+       igc->stopped = 0;
 
        /* Indicate SOL/IDER usage */
        if (igc_check_reset_block(hw) < 0)
@@ -374,6 +914,15 @@ eth_igc_dev_init(struct rte_eth_dev *dev)
                        dev->data->port_id, pci_dev->id.vendor_id,
                        pci_dev->id.device_id);
 
+       rte_intr_callback_register(&pci_dev->intr_handle,
+                       eth_igc_interrupt_handler, (void *)dev);
+
+       /* enable uio/vfio intr/eventfd mapping */
+       rte_intr_enable(&pci_dev->intr_handle);
+
+       /* enable support intr */
+       igc_intr_other_enable(dev);
+
        return 0;
 
 err_late:
@@ -423,16 +972,81 @@ eth_igc_promiscuous_disable(struct rte_eth_dev *dev)
        return 0;
 }
 
+static int
+eth_igc_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
+                      size_t fw_size)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_fw_version fw;
+       int ret;
+
+       igc_get_fw_version(hw, &fw);
+
+       /* if option rom is valid, display its version too */
+       if (fw.or_valid) {
+               ret = snprintf(fw_version, fw_size,
+                        "%d.%d, 0x%08x, %d.%d.%d",
+                        fw.eep_major, fw.eep_minor, fw.etrack_id,
+                        fw.or_major, fw.or_build, fw.or_patch);
+       /* no option rom */
+       } else {
+               if (fw.etrack_id != 0X0000) {
+                       ret = snprintf(fw_version, fw_size,
+                                "%d.%d, 0x%08x",
+                                fw.eep_major, fw.eep_minor,
+                                fw.etrack_id);
+               } else {
+                       ret = snprintf(fw_version, fw_size,
+                                "%d.%d.%d",
+                                fw.eep_major, fw.eep_minor,
+                                fw.eep_build);
+               }
+       }
+
+       ret += 1; /* add the size of '\0' */
+       if (fw_size < (u32)ret)
+               return ret;
+       else
+               return 0;
+}
+
 static int
 eth_igc_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 {
-       PMD_INIT_FUNC_TRACE();
-       RTE_SET_USED(dev);
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
+       dev_info->max_rx_pktlen = MAX_RX_JUMBO_FRAME_SIZE;
+       dev_info->max_mac_addrs = hw->mac.rar_entry_count;
        dev_info->max_rx_queues = IGC_QUEUE_PAIRS_NUM;
        dev_info->max_tx_queues = IGC_QUEUE_PAIRS_NUM;
+       dev_info->max_vmdq_pools = 0;
+
+       dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
+                       ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
+                       ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G;
+
+       dev_info->max_mtu = dev_info->max_rx_pktlen - IGC_ETH_OVERHEAD;
+       dev_info->min_mtu = RTE_ETHER_MIN_MTU;
        return 0;
 }
 
+static int
+eth_igc_led_on(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       return igc_led_on(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
+}
+
+static int
+eth_igc_led_off(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       return igc_led_off(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
+}
+
 static int
 eth_igc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
                uint16_t nb_rx_desc, unsigned int socket_id,
@@ -463,6 +1077,16 @@ eth_igc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
        return 0;
 }
 
+static void eth_igc_tx_queue_release(void *txq)
+{
+       RTE_SET_USED(txq);
+}
+
+static void eth_igc_rx_queue_release(void *rxq)
+{
+       RTE_SET_USED(rxq);
+}
+
 static int
 eth_igc_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
        struct rte_pci_device *pci_dev)
index 4cff0e3..c1e3217 100644 (file)
@@ -18,11 +18,19 @@ extern "C" {
 
 #define IGC_QUEUE_PAIRS_NUM            4
 
+/* structure for interrupt relative data */
+struct igc_interrupt {
+       uint32_t flags;
+       uint32_t mask;
+};
+
 /*
  * Structure to store private data for each driver instance (for each port).
  */
 struct igc_adapter {
        struct igc_hw           hw;
+       struct igc_interrupt    intr;
+       bool            stopped;
 };
 
 #define IGC_DEV_PRIVATE(_dev)  ((_dev)->data->dev_private)
@@ -30,6 +38,33 @@ struct igc_adapter {
 #define IGC_DEV_PRIVATE_HW(_dev) \
        (&((struct igc_adapter *)(_dev)->data->dev_private)->hw)
 
+#define IGC_DEV_PRIVATE_INTR(_dev) \
+       (&((struct igc_adapter *)(_dev)->data->dev_private)->intr)
+
+static inline void
+igc_read_reg_check_set_bits(struct igc_hw *hw, uint32_t reg, uint32_t bits)
+{
+       uint32_t reg_val = IGC_READ_REG(hw, reg);
+
+       bits |= reg_val;
+       if (bits == reg_val)
+               return; /* no need to write back */
+
+       IGC_WRITE_REG(hw, reg, bits);
+}
+
+static inline void
+igc_read_reg_check_clear_bits(struct igc_hw *hw, uint32_t reg, uint32_t bits)
+{
+       uint32_t reg_val = IGC_READ_REG(hw, reg);
+
+       bits = reg_val & ~bits;
+       if (bits == reg_val)
+               return; /* no need to write back */
+
+       IGC_WRITE_REG(hw, reg, bits);
+}
+
 #ifdef __cplusplus
 }
 #endif