Configuring the DPU

Use this information to configure the Napatech SPDK on the DPU for attaching NVMe devices of the target server.

1. On the SoC, load the drivers.

   soc# modprobe virtio_net
   soc# modprobe uio
   soc# rmmod ifc_uio.ko 2>/dev/null
   soc# cd /opt/ipu_workload_nvme/src/
   soc# insmod ipu_mngmnt_tools/software/csc_lek-0.0.0.5/driver/kmod/ifc_uio.ko

2. Initialize the packet processor (PP) switch.

   soc# export CSC_MNGMT_PATH=/opt/ipu_workload_nvme/src/ipu_mngmnt_tools/software/management 
   soc# /opt/ipu_workload_nvme/src/ipu_mngmnt_tools/ppfastpath.sh

   The ppfastpath.sh script contains the following commands.

   cd $CSC_MNGMT_PATH
   ./csc_mngmt --dev 15:00.0 --flush --all
   ./csc_mngmt --dev 15:00.0 --sync
   ./csc_mngmt --dev 15:00.0 --ppcfg fastpath

   An output example:

   PP Fastpath Configuration was SUCCESSFUL!

   Note: If the following error messages are generated, reboot the DPU.

   ** Error: Unable to parse file '/var/csc-config.json'. Error code: 1003. Regenerating...
   *** Error: Invalid Action Group Entry for Sys IF 0 Broadcast whilst trying to configure PP fastpath default config
   *** Error: Invalid Action Group Entry for Sys IF 1 Broadcast whilst trying to configure PP fastpath default config
   *** Error: Invalid Action Group Entry for Sys IF 0 Broadcast whilst trying to configure PP fastpath default config
   *** Error: Invalid Action Group Entry for Sys IF 1 Broadcast whilst trying to configure PP fastpath default config
   PP Fastpath Configuration contained at least one ERROR

   After reboot, the drivers must be loaded again as described in Step 1 and run the ppfastpath.sh script again.

   soc# reboot

3. On the SoC, add destination network interfaces to the PP lookup table.

   soc# cd $CSC_MNGMT_PATH
   soc# ./csc_mngmt  --dev 15:00.0 --lkup --add --lkup-tag 4 \
   --dmac <target_port0_MAC> --route-dest LINE:0:0
   soc# ./csc_mngmt  --dev 15:00.0 --lkup --add --lkup-tag 4 \
   --dmac <target_port1_MAC> --route-dest LINE:1:0

   where:

   • target_port0_MAC is the MAC address of the NIC on the target server, which is connected to port 0 of the DPU.
   • target_port1_MAC is the MAC address of the NIC on the target server, which is connected to port 1 of the DPU.

   For example:

   cd $CSC_MNGMT_PATH
   soc# ./csc_mngmt  --dev 15:00.0 --lkup --add --lkup-tag 4 \
   --dmac 64:9d:99:ff:ed:dc --route-dest LINE:0:0
   soc# ./csc_mngmt  --dev 15:00.0 --lkup --add --lkup-tag 4 \
   --dmac 64:9d:99:ff:ed:dd --route-dest LINE:1:0

4. Set the following network parameter on the SoC to avoid the ARP flux issue.

   soc# echo 1 > /proc/sys/net/ipv4/conf/all/arp_ignore

   This configuration enables responding to the ARP requests only if the target IP is a local address that is configured on the incoming interface.

5. On the SoC, configure IP addresses of two interfaces.

   soc# ip link set dev ens6f1 down
   soc# ip a a 172.168.1.1/24 dev ens6f1
   soc# ip link set dev ens6f1 up
   
   soc# ip link set dev ens6f2 down
   soc# ip a a 172.168.2.1/24 dev ens6f2
   soc# ip link set dev ens6f2 up

   Note: ens6f4 must not be used.
6. On the SoC, verify the network connections between the DPU and the target server.

   For example:

   soc# ping 172.168.1.2
   soc# ping 172.168.2.2

   If it fails to ping to the target server, make sure that:

   • All network connections are established.
   • Pluggable modules are securely inserted.
   • Cables are properly connected.
   • Related devices are functioning correctly without defects.
7. On the SoC, enable the drivers.

   soc# /sbin/rmmod vfio-pci
   soc# /sbin/rmmod vfio_iommu_type1
   soc# /sbin/rmmod vfio
   soc# /sbin/modprobe vfio-pci
   soc# chmod a+x /dev/vfio

8. Bind the management PCI function of the DPU to the vfio-pci driver.

   The PCIe bus ID of the management PCI function on the DPU is 0000:15:00.4.

   soc# cd /opt/ipu_workload_nvme/src/
   soc# ipu_spdk/dpdk/usertools/dpdk-devbind.py -b vfio-pci 0000:15:00.4

9. On the SoC, start bsc_tgt application. Keep the terminal open.

   soc# cd /opt/ipu_workload_nvme/src/
   soc# ipu_spdk/test/external_code/bsc_tgt/bsc_tgt -s 5G \
   --base-virtaddr 0x2000000000 -m 0x0f

   where:

   • -s 5G: Specifies the size of the shared memory used by the application to 5 Gbytes.
   • --base-virtaddr 0x2000000000: Sets the starting address for the application to allocate the memory.
   • -m 0x0f: Configures the CPU affinity mask for the application. The value 0x0f indicates the application to run on CPU cores 0, 1, 2, and 3.

   An output example:

   [2025-01-28 09:53:38.643685] Starting SPDK v22.01.3-pre / DPDK 21.11.2 initialization...
   [2025-01-28 09:53:38.643756] [ DPDK EAL parameters: [2025-01-28 09:53:38.643764] virtio_blk_net [2025-01-28 09:53:38.643769] --no-shconf [2025-01-28 09:53:38.643774] -c 0x0f [2025-01-28 09:53:38.643779] -m 5120 [2025-01-28 09:53:38.643784] --log-level=lib.eal:6 [2025-01-28 09:53:38.643791] --log-level=lib.cryptodev:5 [2025-01-28 09:53:38.643796] --log-level=user1:6 [2025-01-28 09:53:38.643808] --base-virtaddr=0x2000000000 [2025-01-28 09:53:38.643814] --match-allocations [2025-01-28 09:53:38.643820] --file-prefix=spdk_pid1095 [2025-01-28 09:53:38.643826] ]
   EAL: No available 2048 kB hugepages reported
   TELEMETRY: No legacy callbacks, legacy socket not created
   [2025-01-28 09:53:39.359891] app.c: 601:spdk_app_start: *NOTICE*: Total cores available: 4
   [2025-01-28 09:53:39.484647] reactor.c: 943:reactor_run: *NOTICE*: Reactor started on core 1
   [2025-01-28 09:53:39.484833] reactor.c: 943:reactor_run: *NOTICE*: Reactor started on core 2
   [2025-01-28 09:53:39.484987] reactor.c: 943:reactor_run: *NOTICE*: Reactor started on core 3
   [2025-01-28 09:53:39.485048] reactor.c: 943:reactor_run: *NOTICE*: Reactor started on core 0
   [2025-01-28 09:53:39.485245] accel_engine.c: 510:spdk_accel_engine_initialize: *NOTICE*: Accel engine initialized to use software engine.