net/af_packet: run on kernel without qdisc bypass support
[dpdk.git] / drivers / net / ngbe / ngbe_ethdev.c
index aa042f8..3b5c661 100644 (file)
 #include "ngbe_rxtx.h"
 
 static int ngbe_dev_close(struct rte_eth_dev *dev);
-
+static int ngbe_dev_link_update(struct rte_eth_dev *dev,
+                               int wait_to_complete);
+
+static void ngbe_dev_link_status_print(struct rte_eth_dev *dev);
+static int ngbe_dev_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on);
+static int ngbe_dev_macsec_interrupt_setup(struct rte_eth_dev *dev);
+static int ngbe_dev_misc_interrupt_setup(struct rte_eth_dev *dev);
+static int ngbe_dev_rxq_interrupt_setup(struct rte_eth_dev *dev);
 static void ngbe_dev_interrupt_handler(void *param);
 static void ngbe_dev_interrupt_delayed_handler(void *param);
+static void ngbe_configure_msix(struct rte_eth_dev *dev);
 
 /*
  * The set of PCI devices this driver supports
@@ -54,6 +62,25 @@ static const struct rte_eth_desc_lim tx_desc_lim = {
 
 static const struct eth_dev_ops ngbe_eth_dev_ops;
 
+static inline int32_t
+ngbe_pf_reset_hw(struct ngbe_hw *hw)
+{
+       uint32_t ctrl_ext;
+       int32_t status;
+
+       status = hw->mac.reset_hw(hw);
+
+       ctrl_ext = rd32(hw, NGBE_PORTCTL);
+       /* Set PF Reset Done bit so PF/VF Mail Ops can work */
+       ctrl_ext |= NGBE_PORTCTL_RSTDONE;
+       wr32(hw, NGBE_PORTCTL, ctrl_ext);
+       ngbe_flush(hw);
+
+       if (status == NGBE_ERR_SFP_NOT_PRESENT)
+               status = 0;
+       return status;
+}
+
 static inline void
 ngbe_enable_intr(struct rte_eth_dev *dev)
 {
@@ -110,6 +137,8 @@ eth_ngbe_dev_init(struct rte_eth_dev *eth_dev, void *init_params __rte_unused)
        PMD_INIT_FUNC_TRACE();
 
        eth_dev->dev_ops = &ngbe_eth_dev_ops;
+       eth_dev->rx_pkt_burst = &ngbe_recv_pkts;
+       eth_dev->tx_pkt_burst = &ngbe_xmit_pkts_simple;
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
@@ -198,6 +227,13 @@ eth_ngbe_dev_init(struct rte_eth_dev *eth_dev, void *init_params __rte_unused)
        wr32(hw, NGBE_PORTCTL, ctrl_ext);
        ngbe_flush(hw);
 
+       PMD_INIT_LOG(DEBUG, "MAC: %d, PHY: %d",
+                       (int)hw->mac.type, (int)hw->phy.type);
+
+       PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
+                    eth_dev->data->port_id, pci_dev->id.vendor_id,
+                    pci_dev->id.device_id);
+
        rte_intr_callback_register(intr_handle,
                                   ngbe_dev_interrupt_handler, eth_dev);
 
@@ -220,7 +256,7 @@ eth_ngbe_dev_uninit(struct rte_eth_dev *eth_dev)
 
        ngbe_dev_close(eth_dev);
 
-       return -EINVAL;
+       return 0;
 }
 
 static int
@@ -272,17 +308,315 @@ ngbe_dev_configure(struct rte_eth_dev *dev)
        return 0;
 }
 
+static void
+ngbe_dev_phy_intr_setup(struct rte_eth_dev *dev)
+{
+       struct ngbe_hw *hw = ngbe_dev_hw(dev);
+       struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+
+       wr32(hw, NGBE_GPIODIR, NGBE_GPIODIR_DDR(1));
+       wr32(hw, NGBE_GPIOINTEN, NGBE_GPIOINTEN_INT(3));
+       wr32(hw, NGBE_GPIOINTTYPE, NGBE_GPIOINTTYPE_LEVEL(0));
+       if (hw->phy.type == ngbe_phy_yt8521s_sfi)
+               wr32(hw, NGBE_GPIOINTPOL, NGBE_GPIOINTPOL_ACT(0));
+       else
+               wr32(hw, NGBE_GPIOINTPOL, NGBE_GPIOINTPOL_ACT(3));
+
+       intr->mask_misc |= NGBE_ICRMISC_GPIO;
+}
+
+/*
+ * Configure device link speed and setup link.
+ * It returns 0 on success.
+ */
+static int
+ngbe_dev_start(struct rte_eth_dev *dev)
+{
+       struct ngbe_hw *hw = ngbe_dev_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_vector = 0;
+       int err;
+       bool link_up = false, negotiate = false;
+       uint32_t speed = 0;
+       uint32_t allowed_speeds = 0;
+       int status;
+       uint32_t *link_speeds;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* disable uio/vfio intr/eventfd mapping */
+       rte_intr_disable(intr_handle);
+
+       /* stop adapter */
+       hw->adapter_stopped = 0;
+       ngbe_stop_hw(hw);
+
+       /* reinitialize adapter, this calls reset and start */
+       hw->nb_rx_queues = dev->data->nb_rx_queues;
+       hw->nb_tx_queues = dev->data->nb_tx_queues;
+       status = ngbe_pf_reset_hw(hw);
+       if (status != 0)
+               return -1;
+       hw->mac.start_hw(hw);
+       hw->mac.get_link_status = true;
+
+       ngbe_dev_phy_intr_setup(dev);
+
+       /* check and configure queue intr-vector mapping */
+       if ((rte_intr_cap_multiple(intr_handle) ||
+            !RTE_ETH_DEV_SRIOV(dev).active) &&
+           dev->data->dev_conf.intr_conf.rxq != 0) {
+               intr_vector = dev->data->nb_rx_queues;
+               if (rte_intr_efd_enable(intr_handle, intr_vector))
+                       return -1;
+       }
+
+       if (rte_intr_dp_is_en(intr_handle) && intr_handle->intr_vec == NULL) {
+               intr_handle->intr_vec =
+                       rte_zmalloc("intr_vec",
+                                   dev->data->nb_rx_queues * sizeof(int), 0);
+               if (intr_handle->intr_vec == NULL) {
+                       PMD_INIT_LOG(ERR,
+                                    "Failed to allocate %d rx_queues intr_vec",
+                                    dev->data->nb_rx_queues);
+                       return -ENOMEM;
+               }
+       }
+
+       /* confiugre MSI-X for sleep until Rx interrupt */
+       ngbe_configure_msix(dev);
+
+       /* initialize transmission unit */
+       ngbe_dev_tx_init(dev);
+
+       /* This can fail when allocating mbufs for descriptor rings */
+       err = ngbe_dev_rx_init(dev);
+       if (err != 0) {
+               PMD_INIT_LOG(ERR, "Unable to initialize Rx hardware");
+               goto error;
+       }
+
+       err = ngbe_dev_rxtx_start(dev);
+       if (err < 0) {
+               PMD_INIT_LOG(ERR, "Unable to start rxtx queues");
+               goto error;
+       }
+
+       err = hw->mac.check_link(hw, &speed, &link_up, 0);
+       if (err != 0)
+               goto error;
+       dev->data->dev_link.link_status = link_up;
+
+       link_speeds = &dev->data->dev_conf.link_speeds;
+       if (*link_speeds == ETH_LINK_SPEED_AUTONEG)
+               negotiate = true;
+
+       err = hw->mac.get_link_capabilities(hw, &speed, &negotiate);
+       if (err != 0)
+               goto error;
+
+       allowed_speeds = 0;
+       if (hw->mac.default_speeds & NGBE_LINK_SPEED_1GB_FULL)
+               allowed_speeds |= ETH_LINK_SPEED_1G;
+       if (hw->mac.default_speeds & NGBE_LINK_SPEED_100M_FULL)
+               allowed_speeds |= ETH_LINK_SPEED_100M;
+       if (hw->mac.default_speeds & NGBE_LINK_SPEED_10M_FULL)
+               allowed_speeds |= ETH_LINK_SPEED_10M;
+
+       if (*link_speeds & ~allowed_speeds) {
+               PMD_INIT_LOG(ERR, "Invalid link setting");
+               goto error;
+       }
+
+       speed = 0x0;
+       if (*link_speeds == ETH_LINK_SPEED_AUTONEG) {
+               speed = hw->mac.default_speeds;
+       } else {
+               if (*link_speeds & ETH_LINK_SPEED_1G)
+                       speed |= NGBE_LINK_SPEED_1GB_FULL;
+               if (*link_speeds & ETH_LINK_SPEED_100M)
+                       speed |= NGBE_LINK_SPEED_100M_FULL;
+               if (*link_speeds & ETH_LINK_SPEED_10M)
+                       speed |= NGBE_LINK_SPEED_10M_FULL;
+       }
+
+       hw->phy.init_hw(hw);
+       err = hw->mac.setup_link(hw, speed, link_up);
+       if (err != 0)
+               goto error;
+
+       if (rte_intr_allow_others(intr_handle)) {
+               ngbe_dev_misc_interrupt_setup(dev);
+               /* check if lsc interrupt is enabled */
+               if (dev->data->dev_conf.intr_conf.lsc != 0)
+                       ngbe_dev_lsc_interrupt_setup(dev, TRUE);
+               else
+                       ngbe_dev_lsc_interrupt_setup(dev, FALSE);
+               ngbe_dev_macsec_interrupt_setup(dev);
+               ngbe_set_ivar_map(hw, -1, 1, NGBE_MISC_VEC_ID);
+       } else {
+               rte_intr_callback_unregister(intr_handle,
+                                            ngbe_dev_interrupt_handler, dev);
+               if (dev->data->dev_conf.intr_conf.lsc != 0)
+                       PMD_INIT_LOG(INFO,
+                                    "LSC won't enable because of no intr multiplex");
+       }
+
+       /* check if rxq interrupt is enabled */
+       if (dev->data->dev_conf.intr_conf.rxq != 0 &&
+           rte_intr_dp_is_en(intr_handle))
+               ngbe_dev_rxq_interrupt_setup(dev);
+
+       /* enable UIO/VFIO intr/eventfd mapping */
+       rte_intr_enable(intr_handle);
+
+       /* resume enabled intr since HW reset */
+       ngbe_enable_intr(dev);
+
+       if ((hw->sub_system_id & NGBE_OEM_MASK) == NGBE_LY_M88E1512_SFP ||
+               (hw->sub_system_id & NGBE_OEM_MASK) == NGBE_LY_YT8521S_SFP) {
+               /* gpio0 is used to power on/off control*/
+               wr32(hw, NGBE_GPIODATA, 0);
+       }
+
+       /*
+        * Update link status right before return, because it may
+        * start link configuration process in a separate thread.
+        */
+       ngbe_dev_link_update(dev, 0);
+
+       return 0;
+
+error:
+       PMD_INIT_LOG(ERR, "failure in dev start: %d", err);
+       ngbe_dev_clear_queues(dev);
+       return -EIO;
+}
+
+/*
+ * Stop device: disable rx and tx functions to allow for reconfiguring.
+ */
+static int
+ngbe_dev_stop(struct rte_eth_dev *dev)
+{
+       struct rte_eth_link link;
+       struct ngbe_hw *hw = ngbe_dev_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 (hw->adapter_stopped)
+               return 0;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if ((hw->sub_system_id & NGBE_OEM_MASK) == NGBE_LY_M88E1512_SFP ||
+               (hw->sub_system_id & NGBE_OEM_MASK) == NGBE_LY_YT8521S_SFP) {
+               /* gpio0 is used to power on/off control*/
+               wr32(hw, NGBE_GPIODATA, NGBE_GPIOBIT_0);
+       }
+
+       /* disable interrupts */
+       ngbe_disable_intr(hw);
+
+       /* reset the NIC */
+       ngbe_pf_reset_hw(hw);
+       hw->adapter_stopped = 0;
+
+       /* stop adapter */
+       ngbe_stop_hw(hw);
+
+       ngbe_dev_clear_queues(dev);
+
+       /* Clear 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,
+                                          ngbe_dev_interrupt_handler,
+                                          (void *)dev);
+
+       /* Clean datapath event and queue/vec mapping */
+       rte_intr_efd_disable(intr_handle);
+       if (intr_handle->intr_vec != NULL) {
+               rte_free(intr_handle->intr_vec);
+               intr_handle->intr_vec = NULL;
+       }
+
+       hw->adapter_stopped = true;
+       dev->data->dev_started = 0;
+
+       return 0;
+}
+
 /*
  * Reset and stop device.
  */
 static int
 ngbe_dev_close(struct rte_eth_dev *dev)
 {
+       struct ngbe_hw *hw = ngbe_dev_hw(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       int retries = 0;
+       int ret;
+
        PMD_INIT_FUNC_TRACE();
 
-       RTE_SET_USED(dev);
+       ngbe_pf_reset_hw(hw);
+
+       ngbe_dev_stop(dev);
+
+       ngbe_dev_free_queues(dev);
+
+       /* reprogram the RAR[0] in case user changed it. */
+       ngbe_set_rar(hw, 0, hw->mac.addr, 0, true);
+
+       /* Unlock any pending hardware semaphore */
+       ngbe_swfw_lock_reset(hw);
+
+       /* disable uio intr before callback unregister */
+       rte_intr_disable(intr_handle);
+
+       do {
+               ret = rte_intr_callback_unregister(intr_handle,
+                               ngbe_dev_interrupt_handler, dev);
+               if (ret >= 0 || ret == -ENOENT) {
+                       break;
+               } else if (ret != -EAGAIN) {
+                       PMD_INIT_LOG(ERR,
+                               "intr callback unregister failed: %d",
+                               ret);
+               }
+               rte_delay_ms(100);
+       } while (retries++ < (10 + NGBE_LINK_UP_TIME));
+
+       rte_free(dev->data->mac_addrs);
+       dev->data->mac_addrs = NULL;
+
+       rte_free(dev->data->hash_mac_addrs);
+       dev->data->hash_mac_addrs = NULL;
+
+       return ret;
+}
+
+/*
+ * Reset PF device.
+ */
+static int
+ngbe_dev_reset(struct rte_eth_dev *dev)
+{
+       int ret;
+
+       ret = eth_ngbe_dev_uninit(dev);
+       if (ret != 0)
+               return ret;
 
-       return -EINVAL;
+       ret = eth_ngbe_dev_init(dev, NULL);
+
+       return ret;
 }
 
 static int
@@ -292,6 +626,8 @@ ngbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 
        dev_info->max_rx_queues = (uint16_t)hw->mac.max_rx_queues;
        dev_info->max_tx_queues = (uint16_t)hw->mac.max_tx_queues;
+       dev_info->min_rx_bufsize = 1024;
+       dev_info->max_rx_pktlen = 15872;
 
        dev_info->default_rxconf = (struct rte_eth_rxconf) {
                .rx_thresh = {
@@ -321,6 +657,8 @@ ngbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
                                ETH_LINK_SPEED_10M;
 
        /* Driver-preferred Rx/Tx parameters */
+       dev_info->default_rxportconf.burst_size = 32;
+       dev_info->default_txportconf.burst_size = 32;
        dev_info->default_rxportconf.nb_queues = 1;
        dev_info->default_txportconf.nb_queues = 1;
        dev_info->default_rxportconf.ring_size = 256;
@@ -414,6 +752,106 @@ ngbe_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
        return ngbe_dev_link_update_share(dev, wait_to_complete);
 }
 
+/**
+ * It clears the interrupt causes and enables the interrupt.
+ * It will be called once only during NIC initialized.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ * @param on
+ *  Enable or Disable.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+ngbe_dev_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
+{
+       struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+
+       ngbe_dev_link_status_print(dev);
+       if (on != 0) {
+               intr->mask_misc |= NGBE_ICRMISC_PHY;
+               intr->mask_misc |= NGBE_ICRMISC_GPIO;
+       } else {
+               intr->mask_misc &= ~NGBE_ICRMISC_PHY;
+               intr->mask_misc &= ~NGBE_ICRMISC_GPIO;
+       }
+
+       return 0;
+}
+
+/**
+ * It clears the interrupt causes and enables the interrupt.
+ * It will be called once only during NIC initialized.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+ngbe_dev_misc_interrupt_setup(struct rte_eth_dev *dev)
+{
+       struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+       u64 mask;
+
+       mask = NGBE_ICR_MASK;
+       mask &= (1ULL << NGBE_MISC_VEC_ID);
+       intr->mask |= mask;
+       intr->mask_misc |= NGBE_ICRMISC_GPIO;
+
+       return 0;
+}
+
+/**
+ * It clears the interrupt causes and enables the interrupt.
+ * It will be called once only during NIC initialized.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+ngbe_dev_rxq_interrupt_setup(struct rte_eth_dev *dev)
+{
+       struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+       u64 mask;
+
+       mask = NGBE_ICR_MASK;
+       mask &= ~((1ULL << NGBE_RX_VEC_START) - 1);
+       intr->mask |= mask;
+
+       return 0;
+}
+
+/**
+ * It clears the interrupt causes and enables the interrupt.
+ * It will be called once only during NIC initialized.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+ngbe_dev_macsec_interrupt_setup(struct rte_eth_dev *dev)
+{
+       struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+
+       intr->mask_misc |= NGBE_ICRMISC_LNKSEC;
+
+       return 0;
+}
+
 /*
  * It reads ICR and sets flag for the link_update.
  *
@@ -610,10 +1048,110 @@ ngbe_dev_interrupt_handler(void *param)
        ngbe_dev_interrupt_action(dev);
 }
 
+/**
+ * Set the IVAR registers, mapping interrupt causes to vectors
+ * @param hw
+ *  pointer to ngbe_hw struct
+ * @direction
+ *  0 for Rx, 1 for Tx, -1 for other causes
+ * @queue
+ *  queue to map the corresponding interrupt to
+ * @msix_vector
+ *  the vector to map to the corresponding queue
+ */
+void
+ngbe_set_ivar_map(struct ngbe_hw *hw, int8_t direction,
+                  uint8_t queue, uint8_t msix_vector)
+{
+       uint32_t tmp, idx;
+
+       if (direction == -1) {
+               /* other causes */
+               msix_vector |= NGBE_IVARMISC_VLD;
+               idx = 0;
+               tmp = rd32(hw, NGBE_IVARMISC);
+               tmp &= ~(0xFF << idx);
+               tmp |= (msix_vector << idx);
+               wr32(hw, NGBE_IVARMISC, tmp);
+       } else {
+               /* rx or tx causes */
+               /* Workround for ICR lost */
+               idx = ((16 * (queue & 1)) + (8 * direction));
+               tmp = rd32(hw, NGBE_IVAR(queue >> 1));
+               tmp &= ~(0xFF << idx);
+               tmp |= (msix_vector << idx);
+               wr32(hw, NGBE_IVAR(queue >> 1), tmp);
+       }
+}
+
+/**
+ * Sets up the hardware to properly generate MSI-X interrupts
+ * @hw
+ *  board private structure
+ */
+static void
+ngbe_configure_msix(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 ngbe_hw *hw = ngbe_dev_hw(dev);
+       uint32_t queue_id, base = NGBE_MISC_VEC_ID;
+       uint32_t vec = NGBE_MISC_VEC_ID;
+       uint32_t gpie;
+
+       /*
+        * Won't configure MSI-X register if no mapping is done
+        * between intr vector and event fd
+        * but if MSI-X has been enabled already, need to configure
+        * auto clean, auto mask and throttling.
+        */
+       gpie = rd32(hw, NGBE_GPIE);
+       if (!rte_intr_dp_is_en(intr_handle) &&
+           !(gpie & NGBE_GPIE_MSIX))
+               return;
+
+       if (rte_intr_allow_others(intr_handle)) {
+               base = NGBE_RX_VEC_START;
+               vec = base;
+       }
+
+       /* setup GPIE for MSI-X mode */
+       gpie = rd32(hw, NGBE_GPIE);
+       gpie |= NGBE_GPIE_MSIX;
+       wr32(hw, NGBE_GPIE, gpie);
+
+       /* Populate the IVAR table and set the ITR values to the
+        * corresponding register.
+        */
+       if (rte_intr_dp_is_en(intr_handle)) {
+               for (queue_id = 0; queue_id < dev->data->nb_rx_queues;
+                       queue_id++) {
+                       /* by default, 1:1 mapping */
+                       ngbe_set_ivar_map(hw, 0, queue_id, vec);
+                       intr_handle->intr_vec[queue_id] = vec;
+                       if (vec < base + intr_handle->nb_efd - 1)
+                               vec++;
+               }
+
+               ngbe_set_ivar_map(hw, -1, 1, NGBE_MISC_VEC_ID);
+       }
+       wr32(hw, NGBE_ITR(NGBE_MISC_VEC_ID),
+                       NGBE_ITR_IVAL_1G(NGBE_QUEUE_ITR_INTERVAL_DEFAULT)
+                       | NGBE_ITR_WRDSA);
+}
+
 static const struct eth_dev_ops ngbe_eth_dev_ops = {
        .dev_configure              = ngbe_dev_configure,
        .dev_infos_get              = ngbe_dev_info_get,
+       .dev_start                  = ngbe_dev_start,
+       .dev_stop                   = ngbe_dev_stop,
+       .dev_close                  = ngbe_dev_close,
+       .dev_reset                  = ngbe_dev_reset,
        .link_update                = ngbe_dev_link_update,
+       .rx_queue_start             = ngbe_dev_rx_queue_start,
+       .rx_queue_stop              = ngbe_dev_rx_queue_stop,
+       .tx_queue_start             = ngbe_dev_tx_queue_start,
+       .tx_queue_stop              = ngbe_dev_tx_queue_stop,
        .rx_queue_setup             = ngbe_dev_rx_queue_setup,
        .rx_queue_release           = ngbe_dev_rx_queue_release,
        .tx_queue_setup             = ngbe_dev_tx_queue_setup,