document vdpa device type

Signed-off-by: Adrian Moreno <[email protected]>

Author: amorenoz

README.md

@@ -258,6 +258,7 @@ This selector is applicable when "deviceType" is "netDevice"(note: this is defau
 | "ddpProfiles" | N        | A map of device selectors                                      | `string` list Default: `null`                     | "ddpProfiles": ["GTPv1-C/U IPv4/IPv6 payload"]                                                   |
 | "isRdma"      | N        | Mount RDMA resources                                           | `bool`  values `true` or `false` Default: `false` | "isRdma": `true`                                                                                 |
 | "needVhostNet"| N        | Share /dev/vhost-net                                           | `bool`  values `true` or `false` Default: `false` | "needVhostNet": `true`                                                                           |
+| "vdpaType"    | N        | The type of vDPA device (virtio, vhost or `nil`)               | `string` values `vhost` or `virtio` Default: `null` | "vdpaType": "vhost"                                                                           |
 
 
 [//]: # (The tables above generated using: https://ozh.github.io/ascii-tables/)

docs/README.md

@@ -7,3 +7,4 @@ This page contains supplimentary documention that users may find useful for vari
 * [Running RDMA application in Kubernetes](rdma/)
 * [SR-IOV network device plugin with DDP](ddp/)
 * [Using node specific config file for running device plugin DaemonSet](config-file)
+* [Using vDPA devices in Kuberenets](vdpa/)

docs/vdpa/README.md

@@ -0,0 +1,24 @@
+# Using vDPA devices in Kuberenets
+
+vDPA-capable devices can be used to expose an accelerated, yet standard interface to pods. These devices implement the virtIO ring layout that is used for the data-plane.
+The control plane, however, is translated by the vDPA kernel framework which takes care of configuring the virtio devices and virtqueues.
+
+The vDPA kernel framework is composed of a vdpa bus (/sys/bus/vdpa), vdpa devices (/sys/bus/vdpa/devices) and vdpa drivers (/sys/bus/vdpa/drivers). Currently, two vdpa drivers are implemented:
+*  virtio_vdpa: Exposes the device as a virtio-net netdev
+*  virtio_vdpa: Exposes the device as a vhost-vdpa device. This device uses an extension of the vhost-net protocol to allow userspace applications access the rings directly
+
+## Supported NICs:
+* Mellanox ConnectX®-6 DX
+
+## vDPA device creation
+Each NIC might require diffent steps to create vDPA devices on top of the VFs.
+
+## Bind the desired vdpa driver
+The vdpa bus works similar to the pci bus. To unbind a driver from a device, run:
+``echo ${DEV_NAME} > /sys/bus/vdpa/devices/${DEV_NAME}/driver/unbind``
+
+To bind a driver to a device, run:
+``echo ${DEV_NAME} > /sys/bus/vdpa/drivers/${DRIVER_NAME}/bind``
+
+## Priviledges
+IPC_LOCK capability privilege is required for "vhost" mode to be used in a Kubernetes Pod.

docs/vdpa/configMap.yaml

@@ -0,0 +1,30 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: sriovdp-config
+  namespace: kube-system
+data:
+  config.json: |
+    {
+        "resourceList": [{
+            {
+                "resourceName": "vdpa_mlx_virtio",
+                "selectors": {
+                    "vendors": ["15b3"],
+                    "devices": ["101e"],
+                    "drivers": ["mlx5_core"],
+                    "vdpaType": "virtio"
+                }
+            },
+            {
+                "resourceName": "vdpa_mlx_vhost",
+                "selectors": {
+                    "vendors": ["15b3"],
+                    "devices": ["101e"],
+                    "drivers": ["mlx5_core"],
+                    "vdpaType": "vhost"
+                }
+            }
+        ]
+    }
+