net/virtio: setup Rx queue interrupts

This patch mainly allocates structure to store queue/irq mapping,
and configure queue/irq mapping down through PCI ops. It also creates
eventfds for each Rx queue and tell the kernel about the eventfd/intr
binding.

Note: So far, we hard-code 1:1 queue/irq mapping (each rx queue has
one exclusive interrupt), like this:
  vec 0 -> config irq
  vec 1 -> rxq0
  vec 2 -> rxq1
  ...

which means, the "vectors" option of QEMU should be configured with
a value >= N+1 (N is the number of the queue pairs).

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
Tested-by: Lei Yao <lei.a.yao@intel.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>

doc/guides/nics/features/virtio.ini

@@ -5,6 +5,7 @@
 ;
 [Features]
 Link status          = Y
+Rx interrupt         = Y
 Queue start/stop     = Y
 Scattered Rx         = Y
 Promiscuous mode     = Y

doc/guides/nics/features/virtio_vec.ini

@@ -5,6 +5,7 @@
 ;
 [Features]
 Link status          = Y
+Rx interrupt         = Y
 Queue start/stop     = Y
 Promiscuous mode     = Y
 Allmulticast mode    = Y

doc/guides/nics/virtio.rst

@@ -87,6 +87,8 @@ In this release, the virtio PMD driver provides the basic functionality of packe
 
 *   Virtio supports Link State interrupt.
 
+*   Virtio supports Rx interrupt (so far, only support 1:1 mapping for queue/interrupt).
+
 *   Virtio supports software vlan stripping and inserting.
 
 *   Virtio supports using port IO to get PCI resource when uio/igb_uio module is not available.
@@ -274,3 +276,61 @@ Example of using the vector version of the virtio poll mode driver in
 ``testpmd``::
 
    testpmd -c 0x7 -n 4 -- -i --txqflags=0xF01 --rxq=1 --txq=1 --nb-cores=1
+
+
+Interrupt mode
+--------------
+
+.. _virtio_interrupt_mode:
+
+There are three kinds of interrupts from a virtio device over PCI bus: config
+interrupt, Rx interrupts, and Tx interrupts. Config interrupt is used for
+notification of device configuration changes, especially link status (lsc).
+Interrupt mode is translated into Rx interrupts in the context of DPDK.
+
+Prerequisites for Rx interrupts
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+To support Rx interrupts,
+#. Check if guest kernel supports VFIO-NOIOMMU:
+
+    Linux started to support VFIO-NOIOMMU since 4.8.0. Make sure the guest
+    kernel is compiled with:
+
+    .. code-block:: console
+
+        CONFIG_VFIO_NOIOMMU=y
+
+#. Properly set msix vectors when starting VM:
+
+    Enable multi-queue when starting VM, and specify msix vectors in qemu
+    cmdline. (N+1) is the minimum, and (2N+2) is mostly recommended.
+
+    .. code-block:: console
+
+        $(QEMU) ... -device virtio-net-pci,mq=on,vectors=2N+2 ...
+
+#. In VM, insert vfio module in NOIOMMU mode:
+
+    .. code-block:: console
+
+        modprobe vfio enable_unsafe_noiommu_mode=1
+        modprobe vfio-pci
+
+#. In VM, bind the virtio device with vfio-pci:
+
+    .. code-block:: console
+
+        python tools/dpdk-devbind.py -b vfio-pci 00:03.0
+
+Example
+~~~~~~~
+
+Here we use l3fwd-power as an example to show how to get started.
+
+    Example:
+
+    .. code-block:: console
+
+        $ l3fwd-power -c 0x3 -- -p 1 -P --config="(0,0,1)" \
+                                               --no-numa --parse-ptype

doc/guides/rel_notes/release_17_02.rst

@@ -78,6 +78,19 @@ New Features
     it might enter into kernel space to wake up those kthreads if
     necessary).
 
+* **Added virtio Rx interrupt suppprt.**
+
+  This feature enables Rx interrupt mode for virtio pci net devices as
+  binded to VFIO (noiommu mode) and drived by virtio PMD.
+
+  With this feature, virtio PMD can switch between polling mode and
+  interrupt mode, to achieve best performance, and at the same time save
+  power. It can work on both legacy and modern virtio devices. At this mode,
+  each rxq is mapped with an exluded MSIx interrupt.
+
+  See the :ref:`Virtio Interrupt Mode <virtio_interrupt_mode>` documentation
+  for more information.
+
 
 Resolved Issues
 ---------------

drivers/net/virtio/virtio_ethdev.c

@@ -1210,6 +1210,82 @@ rx_func_get(struct rte_eth_dev *eth_dev)
 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
 }
 
+/* Only support 1:1 queue/interrupt mapping so far.
+ * TODO: support n:1 queue/interrupt mapping when there are limited number of
+ * interrupt vectors (<N+1).
+ */
+static int
+virtio_queues_bind_intr(struct rte_eth_dev *dev)
+{
+	uint32_t i;
+	struct virtio_hw *hw = dev->data->dev_private;
+
+	PMD_INIT_LOG(INFO, "queue/interrupt binding\n");
+	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
+		dev->intr_handle->intr_vec[i] = i + 1;
+		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
+						 VIRTIO_MSI_NO_VECTOR) {
+			PMD_DRV_LOG(ERR, "failed to set queue vector");
+			return -EBUSY;
+		}
+	}
+
+	return 0;
+}
+
+static int
+virtio_configure_intr(struct rte_eth_dev *dev)
+{
+	struct virtio_hw *hw = dev->data->dev_private;
+
+	if (!rte_intr_cap_multiple(dev->intr_handle)) {
+		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
+		return -ENOTSUP;
+	}
+
+	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
+		PMD_INIT_LOG(ERR, "Fail to create eventfd");
+		return -1;
+	}
+
+	if (!dev->intr_handle->intr_vec) {
+		dev->intr_handle->intr_vec =
+			rte_zmalloc("intr_vec",
+				    hw->max_queue_pairs * sizeof(int), 0);
+		if (!dev->intr_handle->intr_vec) {
+			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
+				     hw->max_queue_pairs);
+			return -ENOMEM;
+		}
+	}
+
+	/* Re-register callback to update max_intr */
+	rte_intr_callback_unregister(dev->intr_handle,
+				     virtio_interrupt_handler,
+				     dev);
+	rte_intr_callback_register(dev->intr_handle,
+				   virtio_interrupt_handler,
+				   dev);
+
+	/* DO NOT try to remove this! This function will enable msix, or QEMU
+	 * will encounter SIGSEGV when DRIVER_OK is sent.
+	 * And for legacy devices, this should be done before queue/vec binding
+	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
+	 * (22) will be ignored.
+	 */
+	if (rte_intr_enable(dev->intr_handle) < 0) {
+		PMD_DRV_LOG(ERR, "interrupt enable failed");
+		return -1;
+	}
+
+	if (virtio_queues_bind_intr(dev) < 0) {
+		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
+		return -1;
+	}
+
+	return 0;
+}
+
 /* reset device and renegotiate features if needed */
 static int
 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
@@ -1306,6 +1382,14 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
 	ret = virtio_alloc_queues(eth_dev);
 	if (ret < 0)
 		return ret;
+
+	if (eth_dev->data->dev_conf.intr_conf.rxq) {
+		if (virtio_configure_intr(eth_dev) < 0) {
+			PMD_INIT_LOG(ERR, "failed to configure interrupt");
+			return -1;
+		}
+	}
+
 	vtpci_reinit_complete(hw);
 
 	if (pci_dev)