docs/doc/source/node_management/kubernetes/hardware_acceleration_devices/configure-sriov-fec-operator-to-enable-hw-accelerators-for-hosted-vran-containarized-workloads.rst

Configure Intel Wireless FEC Accelerators using SR-IOV FEC operator

This section provides the instructions for installing the operator that provides detailed configurability of Wireless Accelerators on ().

The Operator for Intel Wireless Accelerators supports the following accelerators:

• Intel® vRAN Dedicated Accelerator ACC100.
• Intel® Programmable Acceleration Card N3000.
• Intel® vRAN Dedicated Accelerator ACC200.

• The system has been provisioned and unlocked.

1. Source the platform environment.

   $ source /etc/platform/openrc
   ~(keystone_admin)$

2. Upload the Operator.

   ~(keystone_admin)$ system application-upload /usr/local/share/applications/helm/sriov-fec-operator-<version>.tgz
   +---------------+-------------------------------------+
   | Property      | Value                               |
   +---------------+-------------------------------------+
   | active        | False                               |
   | app_version   | 1.0-1                               |
   | created_at    | 2022-09-29T19:47:29.427225+00:00    |
   | manifest_file | fluxcd-manifests                    |
   | manifest_name | sriov-fec-operator-fluxcd-manifests |
   | name          | sriov-fec-operator                  |
   | progress      | None                                |
   | status        | uploading                           |
   | updated_at    | None                                |
   +---------------+-------------------------------------+

3. Configure a different resource name for devices as desired.

   • To change the resource name for ACC100, use the following command:

     ~(keystone_admin)$ system helm-override-update sriov-fec-operator sriov-fec-operator sriov-fec-system --set env.SRIOV_FEC_ACC100_RESOURCE_NAME=intel_acc100_fec

4. Apply the Operator.

   ~(keystone_admin)$ system application-apply sriov-fec-operator
   +---------------+-------------------------------------+
   | Property      | Value                               |
   +---------------+-------------------------------------+
   | active        | False                               |
   | app_version   | 1.0-1                               |
   | created_at    | 2022-09-29T19:47:29.427225+00:00    |
   | manifest_file | fluxcd-manifests                    |
   | manifest_name | sriov-fec-operator-fluxcd-manifests |
   | name          | sriov-fec-operator                  |
   | progress      | None                                |
   | status        | applying                            |
   | updated_at    | 2022-09-29T19:47:33.599867+00:00    |
   +---------------+-------------------------------------+

   ~(keystone_admin)$ system application-show sriov-fec-operator
   +---------------+-------------------------------------+
   | Property      | Value                               |
   +---------------+-------------------------------------+
   | active        | True                                |
   | app_version   | 1.0-1                               |
   | created_at    | 2022-09-29T19:47:29.427225+00:00    |
   | manifest_file | fluxcd-manifests                    |
   | manifest_name | sriov-fec-operator-fluxcd-manifests |
   | name          | sriov-fec-operator                  |
   | progress      | completed                           |
   | status        | applied                             |
   | updated_at    | 2022-09-29T19:50:27.543655+00:00    |
   +---------------+-------------------------------------+

5. Verify that all the operator pods are up and running.

   $ kubectl get pods -n sriov-fec-system
   NAME                                            READY   STATUS    RESTARTS        AGE
   accelerator-discovery-svh87                     1/1     Running   0               3m26s
   sriov-device-plugin-j54hh                       1/1     Running   0               3m26s
   sriov-fec-controller-manager-77bb5b778b-bjmr8   2/2     Running   0               3m28s
   sriov-fec-daemonset-stnjh                       1/1     Running   0               3m26s

6. List all the nodes in the cluster with accelerators installed.

   $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system
   NAME           CONFIGURED
   controller-0   NotRequested

7. Find the address of the of accelerator device to be configured.

   • ACC100

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-08-25T01:33:35Z"
       generation: 1
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "8298897"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: dcab90d9-2fe2-4769-81b0-fdd54e96e287
     spec:
       physicalFunctions: []
     status:
       conditions:
       - lastTransitionTime: "2022-08-25T01:33:35Z"
           message: ""
           observedGeneration: 1
           reason: NotRequested
           status: "False"
           type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 0d5c
           driver: ""
           maxVirtualFunctions: 16
           pciAddress: "0000:17:00.0"
           vendorID: "8086"
           virtualFunctions: []

   • N3000

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-10-21T18:17:55Z"
       generation: 1
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "1996828"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: 05db8606-8236-4efd-99bb-7b5ca20cd02e
     spec:
       physicalFunctions: []
     status:
       conditions:
       - lastTransitionTime: "2022-10-21T18:17:55Z"
         message: ""
         observedGeneration: 1
         reason: NotRequested
         status: "False"
         type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 0d8f
           driver: ""
           maxVirtualFunctions: 8
           pciAddress: 0000:1c:00.0
           vendorID: "8086"
           virtualFunctions: []

   • ACC200

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-10-21T18:31:41Z"
       generation: 1
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "2144487"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: e4e536fc-a777-4e26-974d-71226d43c4ed
     spec:
       physicalFunctions: []
     status:
       conditions:
       - lastTransitionTime: "2022-10-21T18:31:41Z"
         message: ""
         observedGeneration: 1
         reason: NotRequested
         status: "False"
         type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 57c0
           driver: ""
           maxVirtualFunctions: 16
           pciAddress: 0000:f7:00.0
           vendorID: "8086"
           virtualFunctions: []

8. Apply the device configuration.

   1. ACC100 device configuration.

      • The maximum number of that can be configured for ACC100 is 16 .

      • There are 8 queue groups available which can be allocated to any available operation (4GUL/4GDL/5GUL/5GDL) based on the numQueueGroups parameter.

      • The product of numQueueGroups x numAqsPerGroups x aqDepthLog2 x numVfBundles must be less than 32K.

      • The following example creates 1 , configures ACC100's 8 queue groups; allocating 4 queue groups for 5G Uplink and another 4 queue groups for 5G Downlink.

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:17:00.0
          physicalFunction:
            pfDriver: "vfio-pci"
            vfDriver: "vfio-pci"
            vfAmount: 1
            bbDevConfig:
              acc100:
                # pfMode: false = VF Programming, true = PF Programming
                pfMode: false
                numVfBundles: 1
                maxQueueSize: 1024
                uplink4G:
                  numQueueGroups: 0
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink4G:
                  numQueueGroups: 0
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                uplink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
          drainSkip: true

      • The following example creates 2 , configures ACC100's 8 queue groups; allocating 2 queue groups each for 4G Uplink, 4G downlink, 5G Uplink and 5G downlink.

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:17:00.0
          physicalFunction:
            pfDriver: "vfio-pci"
            vfDriver: "vfio-pci"
            vfAmount: 2
            bbDevConfig:
              acc100:
                # pfMode: false = VF Programming, true = PF Programming
                pfMode: false
                numVfBundles: 2
                maxQueueSize: 1024
                uplink4G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink4G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                uplink5G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink5G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
          drainSkip: true

   2. N3000 device configuration.

      • The maximum number of that can be configured for N3000 is 8 .

      • The maximum number of queues that can be mapped to each VF for uplink or downlink is 32.

      • The following configuration for N3000 creates 1 with 32 queues each for 5G uplink and 5G downlink.

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:1c:00.0
          physicalFunction:
            pfDriver: pci-pf-stub
            vfDriver: vfio-pci
            vfAmount: 1
            bbDevConfig:
              n3000:
                # Network Type: either "FPGA_5GNR" or "FPGA_LTE"
                networkType: "FPGA_5GNR"
                # Pf mode: false = VF Programming, true = PF Programming
                pfMode: false
                flrTimeout: 610
                downlink:
                  bandwidth: 3
                  loadBalance: 128
                  queues:
                    vf0: 32
                    vf1: 0
                    vf2: 0
                    vf3: 0
                    vf4: 0
                    vf5: 0
                    vf6: 0
                    vf7: 0
                uplink:
                  bandwidth: 3
                  loadBalance: 128
                  queues:
                    vf0: 32
                    vf1: 0
                    vf2: 0
                    vf3: 0
                    vf4: 0
                    vf5: 0
                    vf6: 0
                    vf7: 0
          drainSkip: true

      • The following configuration for N3000 creates 2 with 16 queues each, mapping 32 queues with 2 for 5G uplink and another 32 queues with 2 for 5G downlink.

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:1c:00.0
          physicalFunction:
            pfDriver: pci-pf-stub
            vfDriver: vfio-pci
            vfAmount: 2
            bbDevConfig:
              n3000:
                # Network Type: either "FPGA_5GNR" or "FPGA_LTE"
                networkType: "FPGA_5GNR"
                # Pf mode: false = VF Programming, true = PF Programming
                pfMode: false
                flrTimeout: 610
                downlink:
                  bandwidth: 3
                  loadBalance: 128
                  queues:
                    vf0: 16
                    vf1: 16
                    vf2: 0
                    vf3: 0
                    vf4: 0
                    vf5: 0
                    vf6: 0
                    vf7: 0
                uplink:
                  bandwidth: 3
                  loadBalance: 128
                  queues:
                    vf0: 16
                    vf1: 16
                    vf2: 0
                    vf3: 0
                    vf4: 0
                    vf5: 0
                    vf6: 0
                    vf7: 0
          drainSkip: true

   3. ACC200 device configuration.

      • The maximum number of that can be configured for ACC200 is 16 .

      • There are 16 queue groups available which can be allocated to any available operation (4GUL/4GDL/5GUL/5GDL) based on the numQueueGroups parameter.

      • The product of numQueueGroups x numAqsPerGroups x aqDepthLog2 x numVfBundles must be less than 64K.

      • The following configuration creates 1 , configures ACC200's 12 queue groups; allocating 16 queues per for 5G processing engine functions(5GUL/5GDL/FFT).

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:f7:00.0
          physicalFunction:
            pfDriver: vfio-pci
            vfDriver: vfio-pci
            vfAmount: 1
            bbDevConfig:
              acc200:
                # Pf mode: false = VF Programming, true = PF Programming
                pfMode: false
                numVfBundles: 1
                maxQueueSize: 1024
                uplink4G:
                  numQueueGroups: 0
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink4G:
                  numQueueGroups: 0
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                uplink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                qfft:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
          drainSkip: true

      • The following configuration creates 2 , configures ACC200's 16 queue groups; allocating 16 queues per for 4G and 5G processing engine functions(4GUL/4GDL/5GUL/5GDL/FFT).

        apiVersion: sriovfec.intel.com/v2
        kind: SriovFecClusterConfig
        metadata:
          name: config
          namespace: sriov-fec-system
        spec:
          priority: 1
          nodeSelector:
            kubernetes.io/hostname: controller-0
          acceleratorSelector:
            pciAddress: 0000:f7:00.0
          physicalFunction:
            pfDriver: vfio-pci
            vfDriver: vfio-pci
            vfAmount: 2
            bbDevConfig:
              acc200:
                # Pf mode: false = VF Programming, true = PF Programming
                pfMode: false
                numVfBundles: 2
                maxQueueSize: 1024
                uplink4G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink4G:
                  numQueueGroups: 2
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                uplink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                downlink5G:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
                qfft:
                  numQueueGroups: 4
                  numAqsPerGroups: 16
                  aqDepthLog2: 4
          drainSkip: true

   4. The SriovFecClusterConfig must be provided with spec.drainSkip: True to avoid node draining.

   5. Create and apply a SriovFecClusterConfig custom resource using the above examples as templates, setting the parameters nodeSelector:kubernetes.io/hostname and acceleratorSelector:pciAddress to select the desired device and configuring vfAmount and numVfBundles as desired.

      $ kubectl apply -f <sriov-fec-config-file-name>.yaml
      sriovfecclusterconfig.sriovfec.intel.com/config created

      Note

      The vfAmount and numVfBundles in SriovFecClusterConfig must be always equal for ACC100/ACC200.

9. Verify that the configuration is applied.

   • An example of ACC100 status after applying 1 configuration.

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-09-29T19:49:59Z"
       generation: 2
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "2935834"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: 1a39b2a6-7512-4f44-8a64-083df7e480f3
     spec:
       physicalFunctions:
       - bbDevConfig:
           acc100:
             downlink4G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 0
             downlink5G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 4
             maxQueueSize: 1024
             numVfBundles: 1
             pfMode: false
             uplink4G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 0
             uplink5G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 4
         pciAddress: "0000:17:00.0"
         pfDriver: vfio-pci
         vfAmount: 1
         vfDriver: vfio-pci
     status:
       conditions:
       - lastTransitionTime: "2022-09-29T20:33:13Z"
         message: Configured successfully
         observedGeneration: 2
         reason: Succeeded
         status: "True"
         type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 0d5c
           driver: vfio-pci
           maxVirtualFunctions: 16
           pciAddress: "0000:17:00.0"
           vendorID: "8086"
           virtualFunctions:
           - deviceID: 0d5d
             driver: vfio-pci
             pciAddress: "0000:18:00.0"

   • An example of N3000 status after applying 2 configuration.

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-10-21T18:17:55Z"
       generation: 2
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "2011601"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: 05db8606-8236-4efd-99bb-7b5ca20cd02e
     spec:
       drainSkip: true
       physicalFunctions:
       - bbDevConfig:
           n3000:
             downlink:
               bandwidth: 3
               loadBalance: 128
               queues:
                 vf0: 16
                 vf1: 16
                 vf2: 0
                 vf3: 0
                 vf4: 0
                 vf5: 0
                 vf6: 0
                 vf7: 0
             flrTimeout: 610
             networkType: FPGA_5GNR
             uplink:
               bandwidth: 3
               loadBalance: 128
               queues:
                 vf0: 16
                 vf1: 16
                 vf2: 0
                 vf3: 0
                 vf4: 0
                 vf5: 0
                 vf6: 0
                 vf7: 0
         pciAddress: 0000:1c:00.0
         pfDriver: pci-pf-stub
         vfAmount: 2
         vfDriver: vfio-pci
     status:
       conditions:
       - lastTransitionTime: "2022-10-21T19:35:18Z"
         message: Configured successfully
         observedGeneration: 2
         reason: Succeeded
         status: "True"
         type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 0d8f
           driver: pci-pf-stub
           maxVirtualFunctions: 8
           pciAddress: 0000:1c:00.0
           vendorID: "8086"
           virtualFunctions:
           - deviceID: 0d90
             driver: vfio-pci
             pciAddress: 0000:1c:00.1
           - deviceID: 0d90
             driver: vfio-pci
             pciAddress: 0000:1c:00.2

   • An example of ACC200 status after applying 1 configuration.

     $ kubectl get sriovfecnodeconfigs.sriovfec.intel.com -n sriov-fec-system controller-0 -o yaml
     apiVersion: sriovfec.intel.com/v2
     kind: SriovFecNodeConfig
     metadata:
       creationTimestamp: "2022-10-21T18:31:41Z"
       generation: 3
       name: controller-0
       namespace: sriov-fec-system
       resourceVersion: "2159562"
       selfLink: /apis/sriovfec.intel.com/v2/namespaces/sriov-fec-system/sriovfecnodeconfigs/controller-0
       uid: e4e536fc-a777-4e26-974d-71226d43c4ed
     spec:
       drainSkip: true
       physicalFunctions:
       - bbDevConfig:
           acc200:
             downlink4G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 2
             downlink5G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 4
             maxQueueSize: 1024
             numVfBundles: 1
             qfft:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 4
             uplink4G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 2
             uplink5G:
               aqDepthLog2: 4
               numAqsPerGroups: 16
               numQueueGroups: 4
         pciAddress: 0000:f7:00.0
         pfDriver: vfio-pci
         vfAmount: 1
         vfDriver: vfio-pci
     status:
       conditions:
       - lastTransitionTime: "2022-10-21T19:48:26Z"
         message: Configured successfully
         observedGeneration: 3
         reason: Succeeded
         status: "True"
         type: Configured
       inventory:
         sriovAccelerators:
         - deviceID: 57c0
           driver: vfio-pci
           maxVirtualFunctions: 16
           pciAddress: 0000:f7:00.0
           vendorID: "8086"
           virtualFunctions:
           - deviceID: 57c1
             driver: vfio-pci
             pciAddress: 0000:f7:00.1

10. Modify Cluster config.

    1. To further modify device configuration, make desired modifications to the sriov-fec custom resource file and re-apply.

       $ kubectl apply -f <sriov-fec-config-file-name>.yaml
       sriovfecclusterconfig.sriovfec.intel.com/config configured

11. Delete SriovFecClusterConfig.

    $ kubectl delete -f <sriov-fec-config-file-name>.yaml
    sriovfecclusterconfig.sriovfec.intel.com "config" deleted

12. Configure for interface.

    operator also supports vfio-pci driver for interface.

    If the vfio-pci driver is used to bind the interface, then a UUID token must be configured as a VFIO_TOKEN to both and interfaces.

    • For the interface, the VFIO_TOKEN is configured by operator and has the default value of 02bddbbf-bbb0-4d79-886b-91bad3fbb510

    • It is highly recommended to change the default vfio-token when configuring the accelerator in vfio mode (ie., vfio-pci driver for PF interface).

    • The VFIO_TOKEN could be changed by setting SRIOV_FEC_VFIO_TOKEN before application Apply with system helm-override-update.

    • This example sets the SRIOV_FEC_VFIO_TOKEN using uuidgen.

      ~(keystone_admin)$ system helm-override-update sriov-fec-operator sriov-fec-operator sriov-fec-system --set env.SRIOV_FEC_VFIO_TOKEN=`uuidgen`

      Note

      You must configure SRIOV_FEC_VFIO_TOKEN before installing the application. If SRIOV_FEC_VFIO_TOKEN needs to be updated after the application is installed, removing the application will be necessary before proceeding with updating SRIOV_FEC_VFIO_TOKEN and reinstalling the application.

    • For the interface, the same VFIO_TOKEN must be configured by the application. You can get the token using the command system helm-override-show sriov-fec-operator sriov-fec-operator sriov-fec-system.

    • To configure ACC100, N3000 and ACC200 in vfio mode, you should provide sriovFecClusterConfig with spec.physicalFunction.pfDriver: vfio-pci.

13. Switch between Static method configuration and Operator method.

    1. Delete SriovFecClusterConfig.
    2. Remove sriov-fec-operator using the command system application-remove.
    3. Apply the configuration using system host-device-modify, see Enable ACC100/ACC200 Hardware Accelerators for Hosted vRAN Containerized Workloads <enabling-mount-bryce-hw-accelerator-for-hosted-vram-containerized-workloads>.

• See Set Up Pods to Use SRIOV to Access ACC100/ACC200 HW Accelerators <set-up-pods-to-use-sriov>.

• Resource Request: The resource name for configured with operator must be intel.com/intel_fec_acc100 for ACC100, intel.com/intel_fec_5g for N3000 and intel.com/intel_fec_acc200 for ACC200 when requested in a pod spec unless the resource name was modified using the system helm-override-update command.

  • Resource request for ACC100.

    resources:
      requests:
        intel.com/intel_fec_acc100: '16'
      limits:
        intel.com/intel_fec_acc100: '16'

  • Resource request for N3000.

    resources:
      requests:
        intel.com/intel_fec_5g: '2'
      limits:
        intel.com/intel_fec_5g: '2'

  • Resource request for ACC200.

    resources:
      requests:
        intel.com/intel_fec_acc200: '16'
      limits:
        intel.com/intel_fec_acc200: '16'

• vfio-token: (in case of vfio mode)

  An application pod can get the token through a pod environment variable.

  For example, reference the pod spec section for vfio token injection.

  env:
  - name: SRIOV_FEC_VFIO_TOKEN
    value: "02bddbbf-bbb0-4d79-886b-91bad3fbb510"

  Note

  The application can get the existing vfio-token using the command below, if the user updates the custom vfio-token. system helm-override-show sriov-fec-operator sriov-fec-operator sriov-fec-system

  If the vfio-token is available by default, it will not be displayed in the output file.

  Note

  Use the default vfio-token for testing purposes only.

• Run the following command once the application pod is ready to get the address of the allocated device along with the token when applicable.

  • ACC100

    sysadmin@controller-0:~$ kubectl exec -ti app-pod -- env | grep FEC
    PCIDEVICE_INTEL_COM_INTEL_FEC_ACC100=0000:32:00.0
    SRIOV_FEC_VFIO_TOKEN=02bddbbf-bbb0-4d79-886b-91bad3fbb510

  • ACC200

    sysadmin@controller-0:~$ kubectl exec -ti app-pod -- env | grep FEC
    PCIDEVICE_INTEL_COM_INTEL_FEC_ACC200=0000:f7:00.0
    SRIOV_FEC_VFIO_TOKEN=02bddbbf-bbb0-4d79-886b-91bad3fbb510

  • N3000

    sysadmin@controller-0:~$ kubectl exec -ti app-pod -- env | grep FEC
    PCIDEVICE_INTEL_COM_INTEL_FEC_5G=0000:1f:00.0

• Applications that are using when the interface is bound with the vfio-pci driver, should provide the vfio-token to the interface.

  For example, a sample application can provide vfio-vf-token via Environment Abstraction Layer (EAL) parameters. ./test-bbdev.py -e="--vfio-vf-token=$SRIOV_FEC_VFIO_TOKEN -a$PCIDEVICE_INTEL_COM_INTEL_FEC_ACC200"