net/igb: support VF mailbox interrupt for link up/down
authorWenzhuo Lu <wenzhuo.lu@intel.com>
Wed, 1 Jun 2016 01:53:09 +0000 (09:53 +0800)
committerBruce Richardson <bruce.richardson@intel.com>
Mon, 20 Jun 2016 15:21:51 +0000 (17:21 +0200)
When using kernel PF and DPDK VF, when the PF driver finds the link
state changes, up -> down or down -> up, the driver will send a
message to VF by mailbox. This link state change may be
triggered by PHY disconnection/reconnection, user config change
like *ifconfig down/up* or interface parameter, like MTU change.

This patch enables the support of the mailbox interrupt,
so VF driver can receive the message for link up/down.
After VF receives this message, VF port need to be reset to
recover. This needs to be handled by the application so this patch
allows the app to register a reset callback so it can reset the VF port.

Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
doc/guides/rel_notes/release_16_07.rst
drivers/net/e1000/igb_ethdev.c

index 906217e..fb0f5da 100644 (file)
@@ -47,7 +47,7 @@ New Features
   * Dropped specific Xen Dom0 code.
   * Dropped specific anonymous mempool code in testpmd.
 
-* **Added mailbox interrupt support for ixgbe VF.**
+* **Added mailbox interrupt support for ixgbe and igb VFs.**
 
   When the physical NIC link comes down or up, the PF driver will send a
   mailbox message to notify each VF. To handle this link up/down event,
index b822992..b98c827 100644 (file)
 #define E1000_INCVALUE_82576         (16 << IGB_82576_TSYNC_SHIFT)
 #define E1000_TSAUXC_DISABLE_SYSTIME 0x80000000
 
+#define E1000_VTIVAR_MISC                0x01740
+#define E1000_VTIVAR_MISC_MASK           0xFF
+#define E1000_VTIVAR_VALID               0x80
+#define E1000_VTIVAR_MISC_MAILBOX        0
+#define E1000_VTIVAR_MISC_INTR_MASK      0x3
+
 static int  eth_igb_configure(struct rte_eth_dev *dev);
 static int  eth_igb_start(struct rte_eth_dev *dev);
 static void eth_igb_stop(struct rte_eth_dev *dev);
@@ -265,6 +271,9 @@ static void eth_igb_assign_msix_vector(struct e1000_hw *hw, int8_t direction,
 static void eth_igb_write_ivar(struct e1000_hw *hw, uint8_t msix_vector,
                               uint8_t index, uint8_t offset);
 static void eth_igb_configure_msix_intr(struct rte_eth_dev *dev);
+static void eth_igbvf_interrupt_handler(struct rte_intr_handle *handle,
+                                       void *param);
+static void igbvf_mbx_process(struct rte_eth_dev *dev);
 
 /*
  * Define VF Stats MACRO for Non "cleared on read" register
@@ -562,6 +571,41 @@ igb_intr_disable(struct e1000_hw *hw)
        E1000_WRITE_FLUSH(hw);
 }
 
+static inline void
+igbvf_intr_enable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /* only for mailbox */
+       E1000_WRITE_REG(hw, E1000_EIAM, 1 << E1000_VTIVAR_MISC_MAILBOX);
+       E1000_WRITE_REG(hw, E1000_EIAC, 1 << E1000_VTIVAR_MISC_MAILBOX);
+       E1000_WRITE_REG(hw, E1000_EIMS, 1 << E1000_VTIVAR_MISC_MAILBOX);
+       E1000_WRITE_FLUSH(hw);
+}
+
+/* only for mailbox now. If RX/TX needed, should extend this function.  */
+static void
+igbvf_set_ivar_map(struct e1000_hw *hw, uint8_t msix_vector)
+{
+       uint32_t tmp = 0;
+
+       /* mailbox */
+       tmp |= (msix_vector & E1000_VTIVAR_MISC_INTR_MASK);
+       tmp |= E1000_VTIVAR_VALID;
+       E1000_WRITE_REG(hw, E1000_VTIVAR_MISC, tmp);
+}
+
+static void
+eth_igbvf_configure_msix_intr(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /* Configure VF other cause ivar */
+       igbvf_set_ivar_map(hw, E1000_VTIVAR_MISC_MAILBOX);
+}
+
 static inline int32_t
 igb_pf_reset_hw(struct e1000_hw *hw)
 {
@@ -950,6 +994,10 @@ eth_igbvf_dev_init(struct rte_eth_dev *eth_dev)
                     eth_dev->data->port_id, pci_dev->id.vendor_id,
                     pci_dev->id.device_id, "igb_mac_82576_vf");
 
+       rte_intr_callback_register(&pci_dev->intr_handle,
+                                  eth_igbvf_interrupt_handler,
+                                  (void *)eth_dev);
+
        return 0;
 }
 
@@ -958,6 +1006,7 @@ eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
 {
        struct e1000_adapter *adapter =
                E1000_DEV_PRIVATE(eth_dev->data->dev_private);
+       struct rte_pci_device *pci_dev = eth_dev->pci_dev;
 
        PMD_INIT_FUNC_TRACE();
 
@@ -974,6 +1023,12 @@ eth_igbvf_dev_uninit(struct rte_eth_dev *eth_dev)
        rte_free(eth_dev->data->mac_addrs);
        eth_dev->data->mac_addrs = NULL;
 
+       /* disable uio intr before callback unregister */
+       rte_intr_disable(&pci_dev->intr_handle);
+       rte_intr_callback_unregister(&pci_dev->intr_handle,
+                                    eth_igbvf_interrupt_handler,
+                                    (void *)eth_dev);
+
        return 0;
 }
 
@@ -2605,6 +2660,69 @@ eth_igb_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
        eth_igb_interrupt_action(dev);
 }
 
+static int
+eth_igbvf_interrupt_get_status(struct rte_eth_dev *dev)
+{
+       uint32_t eicr;
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+
+       igbvf_intr_disable(hw);
+
+       /* read-on-clear nic registers here */
+       eicr = E1000_READ_REG(hw, E1000_EICR);
+       intr->flags = 0;
+
+       if (eicr == E1000_VTIVAR_MISC_MAILBOX)
+               intr->flags |= E1000_FLAG_MAILBOX;
+
+       return 0;
+}
+
+void igbvf_mbx_process(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_mbx_info *mbx = &hw->mbx;
+       u32 in_msg = 0;
+
+       if (mbx->ops.read(hw, &in_msg, 1, 0))
+               return;
+
+       /* PF reset VF event */
+       if (in_msg == E1000_PF_CONTROL_MSG)
+               _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET);
+}
+
+static int
+eth_igbvf_interrupt_action(struct rte_eth_dev *dev)
+{
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+
+       if (intr->flags & E1000_FLAG_MAILBOX) {
+               igbvf_mbx_process(dev);
+               intr->flags &= ~E1000_FLAG_MAILBOX;
+       }
+
+       igbvf_intr_enable(dev);
+       rte_intr_enable(&dev->pci_dev->intr_handle);
+
+       return 0;
+}
+
+static void
+eth_igbvf_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
+                           void *param)
+{
+       struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
+
+       eth_igbvf_interrupt_get_status(dev);
+       eth_igbvf_interrupt_action(dev);
+}
+
 static int
 eth_igb_led_on(struct rte_eth_dev *dev)
 {
@@ -2876,6 +2994,8 @@ igbvf_dev_start(struct rte_eth_dev *dev)
        struct e1000_adapter *adapter =
                E1000_DEV_PRIVATE(dev->data->dev_private);
        int ret;
+       struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
+       uint32_t intr_vector = 0;
 
        PMD_INIT_FUNC_TRACE();
 
@@ -2895,12 +3015,41 @@ igbvf_dev_start(struct rte_eth_dev *dev)
                return ret;
        }
 
+       /* check and configure queue intr-vector mapping */
+       if (dev->data->dev_conf.intr_conf.rxq != 0) {
+               intr_vector = dev->data->nb_rx_queues;
+               ret = rte_intr_efd_enable(intr_handle, intr_vector);
+               if (ret)
+                       return ret;
+       }
+
+       if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
+               intr_handle->intr_vec =
+                       rte_zmalloc("intr_vec",
+                                   dev->data->nb_rx_queues * sizeof(int), 0);
+               if (!intr_handle->intr_vec) {
+                       PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
+                                    " intr_vec\n", dev->data->nb_rx_queues);
+                       return -ENOMEM;
+               }
+       }
+
+       eth_igbvf_configure_msix_intr(dev);
+
+       /* enable uio/vfio intr/eventfd mapping */
+       rte_intr_enable(intr_handle);
+
+       /* resume enabled intr since hw reset */
+       igbvf_intr_enable(dev);
+
        return 0;
 }
 
 static void
 igbvf_dev_stop(struct rte_eth_dev *dev)
 {
+       struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
+
        PMD_INIT_FUNC_TRACE();
 
        igbvf_stop_adapter(dev);
@@ -2912,6 +3061,16 @@ igbvf_dev_stop(struct rte_eth_dev *dev)
        igbvf_set_vfta_all(dev,0);
 
        igb_dev_clear_queues(dev);
+
+       /* disable intr eventfd mapping */
+       rte_intr_disable(intr_handle);
+
+       /* Clean datapath event and queue/vec mapping */
+       rte_intr_efd_disable(intr_handle);
+       if (intr_handle->intr_vec) {
+               rte_free(intr_handle->intr_vec);
+               intr_handle->intr_vec = NULL;
+       }
 }
 
 static void