net/i40e: fix VFIO interrupt mapping in VF

[dpdk.git] / drivers / net / i40e / i40e_ethdev_vf.c

diff --git a/drivers/net/i40e/i40e_ethdev_vf.c b/drivers/net/i40e/i40e_ethdev_vf.c
index a32c196..193d967 100644 (file)
--- a/drivers/net/i40e/i40e_ethdev_vf.c
+++ b/drivers/net/i40e/i40e_ethdev_vf.c
@@ -1432,7 +1432,6 @@ i40evf_dev_interrupt_handler(void *param)
 
 done:
        i40evf_enable_irq0(hw);
-       rte_intr_enable(dev->intr_handle);
 }
 
 static int
@@ -2040,7 +2039,20 @@ i40evf_dev_start(struct rte_eth_dev *dev)
                goto err_mac;
        }
 
+       /* When a VF port is bound to VFIO-PCI, only miscellaneous interrupt
+        * is mapped to VFIO vector 0 in i40evf_dev_init( ).
+        * If previous VFIO interrupt mapping set in i40evf_dev_init( ) is
+        * not cleared, it will fail when rte_intr_enable( ) tries to map Rx
+        * queue interrupt to other VFIO vectors.
+        * So clear uio/vfio intr/evevnfd first to avoid failure.
+        */
+       if (dev->data->dev_conf.intr_conf.rxq != 0) {
+               rte_intr_disable(intr_handle);
+               rte_intr_enable(intr_handle);
+       }
+
        i40evf_enable_queues_intr(dev);
+
        return 0;
 
 err_mac: