5. Intel(R) FPGA 5GNR FEC Poll Mode Driver

The BBDEV FPGA 5GNR FEC poll mode driver (PMD) supports an FPGA implementation of a VRAN LDPC Encode / Decode 5GNR wireless acceleration function, using Intel’s PCI-e and FPGA based Vista Creek device.

5.1. Features

FPGA 5GNR FEC PMD supports the following features:

  • LDPC Encode in the DL
  • LDPC Decode in the UL
  • 8 VFs per PF (physical device)
  • Maximum of 32 UL queues per VF
  • Maximum of 32 DL queues per VF
  • PCIe Gen-3 x8 Interface
  • MSI-X
  • SR-IOV

FPGA 5GNR FEC PMD supports the following BBDEV capabilities:

  • For the LDPC encode operation:
    • RTE_BBDEV_LDPC_CRC_24B_ATTACH : set to attach CRC24B to CB(s)
    • RTE_BBDEV_LDPC_RATE_MATCH : if set then do not do Rate Match bypass
  • For the LDPC decode operation:
    • RTE_BBDEV_LDPC_CRC_TYPE_24B_CHECK : check CRC24B from CB(s)
    • RTE_BBDEV_LDPC_ITERATION_STOP_ENABLE : disable early termination
    • RTE_BBDEV_LDPC_CRC_TYPE_24B_DROP : drops CRC24B bits appended while decoding
    • RTE_BBDEV_LDPC_HQ_COMBINE_IN_ENABLE : provides an input for HARQ combining
    • RTE_BBDEV_LDPC_HQ_COMBINE_OUT_ENABLE : provides an input for HARQ combining
    • RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_IN_ENABLE : HARQ memory input is internal
    • RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_OUT_ENABLE : HARQ memory output is internal
    • RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_LOOPBACK : loopback data to/from HARQ memory
    • RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_FILLERS : HARQ memory includes the fillers bits

5.2. Limitations

FPGA 5GNR FEC does not support the following:

  • Scatter-Gather function

5.3. Installation

Section 3 of the DPDK manual provides instructions on installing and compiling DPDK. The default set of bbdev compile flags may be found in config/common_base, where for example the flag to build the FPGA 5GNR FEC device, CONFIG_RTE_LIBRTE_PMD_BBDEV_FPGA_5GNR_FEC, is already set.

DPDK requires hugepages to be configured as detailed in section 2 of the DPDK manual. The bbdev test application has been tested with a configuration 40 x 1GB hugepages. The hugepage configuration of a server may be examined using:

grep Huge* /proc/meminfo

5.4. Initialization

When the device first powers up, its PCI Physical Functions (PF) can be listed through this command:

sudo lspci -vd8086:0d8f

The physical and virtual functions are compatible with Linux UIO drivers: vfio and igb_uio. However, in order to work the FPGA 5GNR FEC device firstly needs to be bound to one of these linux drivers through DPDK.

5.4.1. Bind PF UIO driver(s)

Install the DPDK igb_uio driver, bind it with the PF PCI device ID and use lspci to confirm the PF device is under use by igb_uio DPDK UIO driver.

The igb_uio driver may be bound to the PF PCI device using one of three methods:

1. PCI functions (physical or virtual, depending on the use case) can be bound to the UIO driver by repeating this command for every function.

cd <dpdk-top-level-directory>
insmod ./build/kmod/igb_uio.ko
echo "8086 0d8f" > /sys/bus/pci/drivers/igb_uio/new_id
lspci -vd8086:0d8f
  1. Another way to bind PF with DPDK UIO driver is by using the dpdk-devbind.py tool
cd <dpdk-top-level-directory>
./usertools/dpdk-devbind.py -b igb_uio 0000:06:00.0

where the PCI device ID (example: 0000:06:00.0) is obtained using lspci -vd8086:0d8f

  1. A third way to bind is to use dpdk-setup.sh tool
cd <dpdk-top-level-directory>
./usertools/dpdk-setup.sh

select 'Bind Ethernet/Crypto/Baseband device to IGB UIO module'
or
select 'Bind Ethernet/Crypto/Baseband device to VFIO module' depending on driver required
enter PCI device ID
select 'Display current Ethernet/Crypto/Baseband device settings' to confirm binding

In the same way the FPGA 5GNR FEC PF can be bound with vfio, but vfio driver does not support SR-IOV configuration right out of the box, so it will need to be patched.

5.4.2. Enable Virtual Functions

Now, it should be visible in the printouts that PCI PF is under igb_uio control “Kernel driver in use: igb_uio“

To show the number of available VFs on the device, read sriov_totalvfs file..

cat /sys/bus/pci/devices/0000\:<b>\:<d>.<f>/sriov_totalvfs

where 0000\:<b>\:<d>.<f> is the PCI device ID

To enable VFs via igb_uio, echo the number of virtual functions intended to enable to max_vfs file..

echo <num-of-vfs> > /sys/bus/pci/devices/0000\:<b>\:<d>.<f>/max_vfs

Afterwards, all VFs must be bound to appropriate UIO drivers as required, same way it was done with the physical function previously.

Enabling SR-IOV via vfio driver is pretty much the same, except that the file name is different:

echo <num-of-vfs> > /sys/bus/pci/devices/0000\:<b>\:<d>.<f>/sriov_numvfs

5.4.3. Configure the VFs through PF

The PCI virtual functions must be configured before working or getting assigned to VMs/Containers. The configuration involves allocating the number of hardware queues, priorities, load balance, bandwidth and other settings necessary for the device to perform FEC functions.

This configuration needs to be executed at least once after reboot or PCI FLR and can be achieved by using the function fpga_5gnr_fec_configure(), which sets up the parameters defined in fpga_5gnr_fec_conf structure:

struct fpga_5gnr_fec_conf {
    bool pf_mode_en;
    uint8_t vf_ul_queues_number[FPGA_5GNR_FEC_NUM_VFS];
    uint8_t vf_dl_queues_number[FPGA_5GNR_FEC_NUM_VFS];
    uint8_t ul_bandwidth;
    uint8_t dl_bandwidth;
    uint8_t ul_load_balance;
    uint8_t dl_load_balance;
    uint16_t flr_time_out;
};

  • pf_mode_en: identifies whether only PF is to be used, or the VFs. PF and VFs are mutually exclusive and cannot run simultaneously. Set to 1 for PF mode enabled. If PF mode is enabled all queues available in the device are assigned exclusively to PF and 0 queues given to VFs.

  • vf_*l_queues_number: defines the hardware queue mapping for every VF.

  • *l_bandwidth: in case of congestion on PCIe interface. The device allocates different bandwidth to UL and DL. The weight is configured by this setting. The unit of weight is 3 code blocks. For example, if the code block cbps (code block per second) ratio between UL and DL is 12:1, then the configuration value should be set to 36:3. The schedule algorithm is based on code block regardless the length of each block.

  • *l_load_balance: hardware queues are load-balanced in a round-robin fashion. Queues get filled first-in first-out until they reach a pre-defined watermark level, if exceeded, they won’t get assigned new code blocks.. This watermark is defined by this setting.

    If all hardware queues exceeds the watermark, no code blocks will be streamed in from UL/DL code block FIFO.

  • flr_time_out: specifies how many 16.384us to be FLR time out. The time_out = flr_time_out x 16.384us. For instance, if you want to set 10ms for the FLR time out then set this setting to 0x262=610.

An example configuration code calling the function fpga_5gnr_fec_configure() is shown below:

struct fpga_5gnr_fec_conf conf;
unsigned int i;

memset(&conf, 0, sizeof(struct fpga_5gnr_fec_conf));
conf.pf_mode_en = 1;

for (i = 0; i < FPGA_5GNR_FEC_NUM_VFS; ++i) {
    conf.vf_ul_queues_number[i] = 4;
    conf.vf_dl_queues_number[i] = 4;
}
conf.ul_bandwidth = 12;
conf.dl_bandwidth = 5;
conf.dl_load_balance = 64;
conf.ul_load_balance = 64;

/* setup FPGA PF */
ret = fpga_5gnr_fec_configure(info->dev_name, &conf);
TEST_ASSERT_SUCCESS(ret,
    "Failed to configure 4G FPGA PF for bbdev %s",
    info->dev_name);

5.5. Test Application

BBDEV provides a test application, test-bbdev.py and range of test data for testing the functionality of FPGA 5GNR FEC encode and decode, depending on the device’s capabilities. The test application is located under app->test-bbdev folder and has the following options:

"-p", "--testapp-path": specifies path to the bbdev test app.
"-e", "--eal-params"  : EAL arguments which are passed to the test app.
"-t", "--timeout"     : Timeout in seconds (default=300).
"-c", "--test-cases"  : Defines test cases to run. Run all if not specified.
"-v", "--test-vector" : Test vector path (default=dpdk_path+/app/test-bbdev/test_vectors/bbdev_null.data).
"-n", "--num-ops"     : Number of operations to process on device (default=32).
"-b", "--burst-size"  : Operations enqueue/dequeue burst size (default=32).
"-l", "--num-lcores"  : Number of lcores to run (default=16).
"-i", "--init-device" : Initialise PF device with default values.

To execute the test application tool using simple decode or encode data, type one of the following:

./test-bbdev.py -c validation -n 64 -b 1 -v ./ldpc_dec_default.data
./test-bbdev.py -c validation -n 64 -b 1 -v ./ldpc_enc_default.data

The test application test-bbdev.py, supports the ability to configure the PF device with a default set of values, if the “-i” or “- -init-device” option is included. The default values are defined in test_bbdev_perf.c as:

  • VF_UL_QUEUE_VALUE 4
  • VF_DL_QUEUE_VALUE 4
  • UL_BANDWIDTH 3
  • DL_BANDWIDTH 3
  • UL_LOAD_BALANCE 128
  • DL_LOAD_BALANCE 128
  • FLR_TIMEOUT 610

5.5.1. Test Vectors

In addition to the simple LDPC decoder and LDPC encoder tests, bbdev also provides a range of additional tests under the test_vectors folder, which may be useful. The results of these tests will depend on the FPGA 5GNR FEC capabilities.