baseband/fpga_lte_fec: add driver for FEC on FPGA
authorNicolas Chautru <nicolas.chautru@intel.com>
Fri, 14 Jun 2019 16:17:36 +0000 (09:17 -0700)
committerAkhil Goyal <akhil.goyal@nxp.com>
Fri, 5 Jul 2019 13:28:14 +0000 (15:28 +0200)
Supports for FEC 4G PMD Driver on FPGA card PAC N3000

Signed-off-by: Nicolas Chautru <nicolas.chautru@intel.com>
Acked-by: Kamil Chalupnik <kamilx.chalupnik@intel.com>
12 files changed:
config/common_base
doc/guides/bbdevs/fpga_lte_fec.rst [new file with mode: 0644]
doc/guides/bbdevs/index.rst
doc/guides/rel_notes/release_19_08.rst
drivers/baseband/Makefile
drivers/baseband/fpga_lte_fec/Makefile [new file with mode: 0644]
drivers/baseband/fpga_lte_fec/fpga_lte_fec.c [new file with mode: 0644]
drivers/baseband/fpga_lte_fec/fpga_lte_fec.h [new file with mode: 0644]
drivers/baseband/fpga_lte_fec/meson.build [new file with mode: 0644]
drivers/baseband/fpga_lte_fec/rte_pmd_bbdev_fpga_lte_fec_version.map [new file with mode: 0644]
drivers/baseband/meson.build
mk/rte.app.mk

index 757f365..a104b5f 100644 (file)
@@ -537,6 +537,7 @@ CONFIG_RTE_PMD_PACKET_PREFETCH=y
 # EXPERIMENTAL: API may change without prior notice
 #
 CONFIG_RTE_LIBRTE_BBDEV=y
+CONFIG_RTE_LIBRTE_BBDEV_DEBUG=n
 CONFIG_RTE_BBDEV_MAX_DEVS=128
 CONFIG_RTE_BBDEV_OFFLOAD_COST=y
 CONFIG_RTE_BBDEV_SDK_AVX2=n
@@ -551,6 +552,11 @@ CONFIG_RTE_LIBRTE_PMD_BBDEV_NULL=y
 #
 CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW=y
 
+#
+# Compile PMD for Intel FPGA LTE FEC bbdev device
+#
+CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC=y
+
 #
 # Compile generic crypto device library
 #
diff --git a/doc/guides/bbdevs/fpga_lte_fec.rst b/doc/guides/bbdevs/fpga_lte_fec.rst
new file mode 100644 (file)
index 0000000..4a3061a
--- /dev/null
@@ -0,0 +1,316 @@
+..  SPDX-License-Identifier: BSD-3-Clause
+    Copyright(c) 2019 Intel Corporation
+
+Intel(R) FPGA LTE FEC Poll Mode Driver
+======================================
+
+The BBDEV FPGA LTE FEC poll mode driver (PMD) supports an FPGA implementation of a VRAN
+Turbo Encode / Decode LTE wireless acceleration function, using Intel's PCI-e and FPGA
+based Vista Creek device.
+
+Features
+--------
+
+FPGA LTE FEC PMD supports the following features:
+
+- Turbo Encode in the DL with total throughput of 4.5 Gbits/s
+- Turbo Decode in the UL with total throughput of 1.5 Gbits/s assuming 8 decoder iterations
+- 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 LTE FEC PMD supports the following BBDEV capabilities:
+
+* For the turbo encode operation:
+   - ``RTE_BBDEV_TURBO_CRC_24B_ATTACH`` :  set to attach CRC24B to CB(s)
+   - ``RTE_BBDEV_TURBO_RATE_MATCH`` :  if set then do not do Rate Match bypass
+   - ``RTE_BBDEV_TURBO_ENC_INTERRUPTS`` :  set for encoder dequeue interrupts
+
+
+* For the turbo decode operation:
+   - ``RTE_BBDEV_TURBO_CRC_TYPE_24B`` :  check CRC24B from CB(s)
+   - ``RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE`` :  perform subblock de-interleave
+   - ``RTE_BBDEV_TURBO_DEC_INTERRUPTS`` :  set for decoder dequeue interrupts
+   - ``RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN`` :  set if negative LLR encoder i/p is supported
+   - ``RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP`` :  keep CRC24B bits appended while decoding
+
+
+Limitations
+-----------
+
+FPGA LTE FEC does not support the following:
+
+- Scatter-Gather function
+
+
+Installation
+--------------
+
+Section 3 of the DPDK manual provides instuctions 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 LTE FEC device, ``CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC``, is already
+set. It is assumed DPDK has been compiled using for instance:
+
+.. code-block:: console
+
+  make install T=x86_64-native-linuxapp-gcc
+
+
+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:
+
+.. code-block:: console
+
+   grep Huge* /proc/meminfo
+
+
+Initialization
+--------------
+
+When the device first powers up, its PCI Physical Functions (PF) can be listed through this command:
+
+.. code-block:: console
+
+  sudo lspci -vd1172:5052
+
+The physical and virtual functions are compatible with Linux UIO drivers:
+``vfio`` and ``igb_uio``. However, in order to work the FPGA LTE FEC device firstly needs
+to be bound to one of these linux drivers through DPDK.
+
+
+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.
+
+.. code-block:: console
+
+  cd <dpdk-top-level-directory>
+  insmod ./build/kmod/igb_uio.ko
+  echo "1172 5052" > /sys/bus/pci/drivers/igb_uio/new_id
+  lspci -vd1172:
+
+
+2. Another way to bind PF with DPDK UIO driver is by using the ``dpdk-devbind.py`` tool
+
+.. code-block:: console
+
+  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 -vd1172:
+
+
+3. A third way to bind is to use ``dpdk-setup.sh`` tool
+
+.. code-block:: console
+
+  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 LTE 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.
+
+
+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..
+
+.. code-block:: console
+
+  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..
+
+.. code-block:: console
+
+  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:
+
+.. code-block:: console
+
+  echo <num-of-vfs> > /sys/bus/pci/devices/0000\:<b>\:<d>.<f>/sriov_numvfs
+
+
+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_lte_fec_configure()``, which sets up the
+parameters defined in ``fpga_lte_fec_conf`` structure:
+
+.. code-block:: c
+
+  struct fpga_lte_fec_conf {
+      bool pf_mode_en;
+      uint8_t vf_ul_queues_number[FPGA_LTE_FEC_NUM_VFS];
+      uint8_t vf_dl_queues_number[FPGA_LTE_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_lte_fec_configure()`` is shown
+below:
+
+.. code-block:: c
+
+  struct fpga_lte_fec_conf conf;
+  unsigned int i;
+
+  memset(&conf, 0, sizeof(struct fpga_lte_fec_conf));
+  conf.pf_mode_en = 1;
+
+  for (i = 0; i < FPGA_LTE_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_lte_fec_configure(info->dev_name, &conf);
+  TEST_ASSERT_SUCCESS(ret,
+      "Failed to configure 4G FPGA PF for bbdev %s",
+      info->dev_name);
+
+
+Test Application
+----------------
+
+BBDEV provides a test application, ``test-bbdev.py`` and range of test data for testing
+the functionality of FPGA LTE FEC turbo encode and turbo decode, depending on the device's
+capabilities. The test application is located under app->test-bbdev folder and has the
+following options:
+
+.. code-block:: console
+
+  "-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 turbo decode or turbo encode data,
+type one of the following:
+
+.. code-block:: console
+
+  ./test-bbdev.py -c validation -n 64 -b 8 -v ./turbo_dec_default.data
+  ./test-bbdev.py -c validation -n 64 -b 8 -v ./turbo_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
+
+
+Test Vectors
+~~~~~~~~~~~~
+
+In addition to the simple turbo decoder and turbo 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 LTE FEC capabilities:
+
+* turbo decoder tests:
+   - ``turbo_dec_c1_k6144_r0_e10376_crc24b_sbd_negllr_high_snr.data``
+   - ``turbo_dec_c1_k6144_r0_e10376_crc24b_sbd_negllr_low_snr.data``
+   - ``turbo_dec_c1_k6144_r0_e34560_negllr.data``
+   - ``turbo_dec_c1_k6144_r0_e34560_sbd_negllr.data``
+   - ``turbo_dec_c2_k3136_r0_e4920_sbd_negllr_crc24b.data``
+   - ``turbo_dec_c2_k3136_r0_e4920_sbd_negllr.data``
+
+
+* turbo encoder tests:
+   - ``turbo_enc_c1_k40_r0_e1190_rm.data``
+   - ``turbo_enc_c1_k40_r0_e1194_rm.data``
+   - ``turbo_enc_c1_k40_r0_e1196_rm.data``
+   - ``turbo_enc_c1_k40_r0_e272_rm.data``
+   - ``turbo_enc_c1_k6144_r0_e18444.data``
+   - ``turbo_enc_c1_k6144_r0_e32256_crc24b_rm.data``
+   - ``turbo_enc_c2_k5952_r0_e17868_crc24b.data``
+   - ``turbo_enc_c3_k4800_r2_e14412_crc24b.data``
+   - ``turbo_enc_c4_k4800_r2_e14412_crc24b.data``
index 93276ed..005b95e 100644 (file)
@@ -10,3 +10,4 @@ Baseband Device Drivers
 
     null
     turbo_sw
+    fpga_lte_fec
index 40d3079..fd3b4b6 100644 (file)
@@ -133,6 +133,12 @@ New Features
   device to be received repeatedly at a high rate. This can be useful for quick
   performance testing of DPDK apps.
 
+* **Added a FPGA_LTE_FEC bbdev PMD.**
+
+  Added the new ``fpga_lte_fec`` bbdev driver for the IntelĀ® FPGA PAC
+  (Programmable  Acceleration Card) N3000.  See the
+  :doc:`../bbdevs/fpga_lte_fec` BBDEV guide for more details on this new driver.
+
 * **Added Intel QuickData Technology PMD**
 
   The PMD for Intel\ |reg|  QuickData Technology, part of
index 4ec83b0..ceffc7d 100644 (file)
@@ -10,5 +10,7 @@ DIRS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_NULL) += null
 DEPDIRS-null = $(core-libs)
 DIRS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += turbo_sw
 DEPDIRS-turbo_sw = $(core-libs)
+DIRS-$(CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC) += fpga_lte_fec
+DEPDIRS-fpga_lte_fec = $(core-libs)
 
 include $(RTE_SDK)/mk/rte.subdir.mk
diff --git a/drivers/baseband/fpga_lte_fec/Makefile b/drivers/baseband/fpga_lte_fec/Makefile
new file mode 100644 (file)
index 0000000..a38a396
--- /dev/null
@@ -0,0 +1,29 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2019 Intel Corporation
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+# library name
+LIB = librte_pmd_fpga_lte_fec.a
+
+# build flags
+CFLAGS += -DALLOW_EXPERIMENTAL_API
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
+LDLIBS += -lrte_bbdev
+LDLIBS += -lrte_pci -lrte_bus_pci
+
+# versioning export map
+EXPORT_MAP := rte_pmd_bbdev_fpga_lte_fec_version.map
+
+# library version
+LIBABIVER := 1
+
+# library source files
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC) += fpga_lte_fec.c
+
+# export include files
+SYMLINK-$(CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC)-include += fpga_lte_fec.h
+
+include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/baseband/fpga_lte_fec/fpga_lte_fec.c b/drivers/baseband/fpga_lte_fec/fpga_lte_fec.c
new file mode 100644 (file)
index 0000000..19e7689
--- /dev/null
@@ -0,0 +1,2674 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Intel Corporation
+ */
+
+#include <unistd.h>
+
+#include <rte_common.h>
+#include <rte_log.h>
+#include <rte_dev.h>
+#include <rte_malloc.h>
+#include <rte_mempool.h>
+#include <rte_errno.h>
+#include <rte_pci.h>
+#include <rte_bus_pci.h>
+#include <rte_byteorder.h>
+#ifdef RTE_BBDEV_OFFLOAD_COST
+#include <rte_cycles.h>
+#endif
+
+#include <rte_bbdev.h>
+#include <rte_bbdev_pmd.h>
+
+#include "fpga_lte_fec.h"
+
+/* Turbo SW PMD logging ID */
+static int fpga_lte_fec_logtype;
+
+/* Helper macro for logging */
+#define rte_bbdev_log(level, fmt, ...) \
+       rte_log(RTE_LOG_ ## level, fpga_lte_fec_logtype, fmt "\n", \
+               ##__VA_ARGS__)
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+#define rte_bbdev_log_debug(fmt, ...) \
+               rte_bbdev_log(DEBUG, "fpga_lte_fec: " fmt, \
+               ##__VA_ARGS__)
+#else
+#define rte_bbdev_log_debug(fmt, ...)
+#endif
+
+/* FPGA LTE FEC driver names */
+#define FPGA_LTE_FEC_PF_DRIVER_NAME intel_fpga_lte_fec_pf
+#define FPGA_LTE_FEC_VF_DRIVER_NAME intel_fpga_lte_fec_vf
+
+/* FPGA LTE FEC PCI vendor & device IDs */
+#define FPGA_LTE_FEC_VENDOR_ID (0x1172)
+#define FPGA_LTE_FEC_PF_DEVICE_ID (0x5052)
+#define FPGA_LTE_FEC_VF_DEVICE_ID (0x5050)
+
+/* Align DMA descriptors to 256 bytes - cache-aligned */
+#define FPGA_RING_DESC_ENTRY_LENGTH (8)
+/* Ring size is in 256 bits (32 bytes) units */
+#define FPGA_RING_DESC_LEN_UNIT_BYTES (32)
+/* Maximum size of queue */
+#define FPGA_RING_MAX_SIZE (1024)
+#define FPGA_FLR_TIMEOUT_UNIT (16.384)
+
+#define FPGA_NUM_UL_QUEUES (32)
+#define FPGA_NUM_DL_QUEUES (32)
+#define FPGA_TOTAL_NUM_QUEUES (FPGA_NUM_UL_QUEUES + FPGA_NUM_DL_QUEUES)
+#define FPGA_NUM_INTR_VEC (FPGA_TOTAL_NUM_QUEUES - RTE_INTR_VEC_RXTX_OFFSET)
+
+#define FPGA_INVALID_HW_QUEUE_ID (0xFFFFFFFF)
+
+#define FPGA_QUEUE_FLUSH_TIMEOUT_US (1000)
+#define FPGA_TIMEOUT_CHECK_INTERVAL (5)
+
+/* FPGA LTE FEC Register mapping on BAR0 */
+enum {
+       FPGA_LTE_FEC_VERSION_ID = 0x00000000, /* len: 4B */
+       FPGA_LTE_FEC_CONFIGURATION = 0x00000004, /* len: 2B */
+       FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE = 0x00000008, /* len: 1B */
+       FPGA_LTE_FEC_LOAD_BALANCE_FACTOR = 0x0000000a, /* len: 2B */
+       FPGA_LTE_FEC_RING_DESC_LEN = 0x0000000c, /* len: 2B */
+       FPGA_LTE_FEC_FLR_TIME_OUT = 0x0000000e, /* len: 2B */
+       FPGA_LTE_FEC_VFQ_FLUSH_STATUS_LW = 0x00000018, /* len: 4B */
+       FPGA_LTE_FEC_VFQ_FLUSH_STATUS_HI = 0x0000001c, /* len: 4B */
+       FPGA_LTE_FEC_VF0_DEBUG = 0x00000020, /* len: 4B */
+       FPGA_LTE_FEC_VF1_DEBUG = 0x00000024, /* len: 4B */
+       FPGA_LTE_FEC_VF2_DEBUG = 0x00000028, /* len: 4B */
+       FPGA_LTE_FEC_VF3_DEBUG = 0x0000002c, /* len: 4B */
+       FPGA_LTE_FEC_VF4_DEBUG = 0x00000030, /* len: 4B */
+       FPGA_LTE_FEC_VF5_DEBUG = 0x00000034, /* len: 4B */
+       FPGA_LTE_FEC_VF6_DEBUG = 0x00000038, /* len: 4B */
+       FPGA_LTE_FEC_VF7_DEBUG = 0x0000003c, /* len: 4B */
+       FPGA_LTE_FEC_QUEUE_MAP = 0x00000040, /* len: 256B */
+       FPGA_LTE_FEC_RING_CTRL_REGS = 0x00000200  /* len: 2048B */
+};
+
+/* FPGA LTE FEC Ring Control Registers */
+enum {
+       FPGA_LTE_FEC_RING_HEAD_ADDR = 0x00000008,
+       FPGA_LTE_FEC_RING_SIZE = 0x00000010,
+       FPGA_LTE_FEC_RING_MISC = 0x00000014,
+       FPGA_LTE_FEC_RING_ENABLE = 0x00000015,
+       FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN = 0x00000016,
+       FPGA_LTE_FEC_RING_SHADOW_TAIL = 0x00000018,
+       FPGA_LTE_FEC_RING_HEAD_POINT = 0x0000001C
+};
+
+/* FPGA LTE FEC DESCRIPTOR ERROR */
+enum {
+       DESC_ERR_NO_ERR = 0x0,
+       DESC_ERR_K_OUT_OF_RANGE = 0x1,
+       DESC_ERR_K_NOT_NORMAL = 0x2,
+       DESC_ERR_KPAI_NOT_NORMAL = 0x3,
+       DESC_ERR_DESC_OFFSET_ERR = 0x4,
+       DESC_ERR_DESC_READ_FAIL = 0x8,
+       DESC_ERR_DESC_READ_TIMEOUT = 0x9,
+       DESC_ERR_DESC_READ_TLP_POISONED = 0xA,
+       DESC_ERR_CB_READ_FAIL = 0xC,
+       DESC_ERR_CB_READ_TIMEOUT = 0xD,
+       DESC_ERR_CB_READ_TLP_POISONED = 0xE
+};
+
+/* FPGA LTE FEC DMA Encoding Request Descriptor */
+struct __attribute__((__packed__)) fpga_dma_enc_desc {
+       uint32_t done:1,
+               rsrvd0:11,
+               error:4,
+               rsrvd1:16;
+       uint32_t ncb:16,
+               rsrvd2:14,
+               rv:2;
+       uint32_t bypass_rm:1,
+               irq_en:1,
+               crc_en:1,
+               rsrvd3:13,
+               offset:10,
+               rsrvd4:6;
+       uint16_t e;
+       uint16_t k;
+       uint32_t out_addr_lw;
+       uint32_t out_addr_hi;
+       uint32_t in_addr_lw;
+       uint32_t in_addr_hi;
+
+       union {
+               struct {
+                       /* Virtual addresses used to retrieve SW context info */
+                       void *op_addr;
+                       /* Stores information about total number of Code Blocks
+                        * in currently processed Transport Block
+                        */
+                       uint64_t cbs_in_op;
+               };
+
+               uint8_t sw_ctxt[FPGA_RING_DESC_LEN_UNIT_BYTES *
+                                       (FPGA_RING_DESC_ENTRY_LENGTH - 1)];
+       };
+};
+
+/* FPGA LTE FEC DMA Decoding Request Descriptor */
+struct __attribute__((__packed__)) fpga_dma_dec_desc {
+       uint32_t done:1,
+               iter:5,
+               rsrvd0:2,
+               crc_pass:1,
+               rsrvd1:3,
+               error:4,
+               crc_type:1,
+               rsrvd2:7,
+               max_iter:5,
+               rsrvd3:3;
+       uint32_t rsrvd4;
+       uint32_t bypass_rm:1,
+               irq_en:1,
+               drop_crc:1,
+               rsrvd5:13,
+               offset:10,
+               rsrvd6:6;
+       uint16_t k;
+       uint16_t in_len;
+       uint32_t out_addr_lw;
+       uint32_t out_addr_hi;
+       uint32_t in_addr_lw;
+       uint32_t in_addr_hi;
+
+       union {
+               struct {
+                       /* Virtual addresses used to retrieve SW context info */
+                       void *op_addr;
+                       /* Stores information about total number of Code Blocks
+                        * in currently processed Transport Block
+                        */
+                       uint8_t cbs_in_op;
+               };
+
+               uint32_t sw_ctxt[8 * (FPGA_RING_DESC_ENTRY_LENGTH - 1)];
+       };
+};
+
+/* FPGA LTE DMA Descriptor */
+union fpga_dma_desc {
+       struct fpga_dma_enc_desc enc_req;
+       struct fpga_dma_dec_desc dec_req;
+};
+
+/* FPGA LTE FEC Ring Control Register */
+struct __attribute__((__packed__)) fpga_ring_ctrl_reg {
+       uint64_t ring_base_addr;
+       uint64_t ring_head_addr;
+       uint16_t ring_size:11;
+       uint16_t rsrvd0;
+       union { /* Miscellaneous register */
+               uint8_t misc;
+               uint8_t max_ul_dec:5,
+                       max_ul_dec_en:1,
+                       rsrvd1:2;
+       };
+       uint8_t enable;
+       uint8_t flush_queue_en;
+       uint8_t rsrvd2;
+       uint16_t shadow_tail;
+       uint16_t rsrvd3;
+       uint16_t head_point;
+       uint16_t rsrvd4;
+
+};
+
+/* Private data structure for each FPGA FEC device */
+struct fpga_lte_fec_device {
+       /** Base address of MMIO registers (BAR0) */
+       void *mmio_base;
+       /** Base address of memory for sw rings */
+       void *sw_rings;
+       /** Physical address of sw_rings */
+       rte_iova_t sw_rings_phys;
+       /** Number of bytes available for each queue in device. */
+       uint32_t sw_ring_size;
+       /** Max number of entries available for each queue in device */
+       uint32_t sw_ring_max_depth;
+       /** Base address of response tail pointer buffer */
+       uint32_t *tail_ptrs;
+       /** Physical address of tail pointers */
+       rte_iova_t tail_ptr_phys;
+       /** Queues flush completion flag */
+       uint64_t *flush_queue_status;
+       /* Bitmap capturing which Queues are bound to the PF/VF */
+       uint64_t q_bound_bit_map;
+       /* Bitmap capturing which Queues have already been assigned */
+       uint64_t q_assigned_bit_map;
+       /** True if this is a PF FPGA FEC device */
+       bool pf_device;
+};
+
+/* Structure associated with each queue. */
+struct __rte_cache_aligned fpga_queue {
+       struct fpga_ring_ctrl_reg ring_ctrl_reg;  /* Ring Control Register */
+       union fpga_dma_desc *ring_addr;  /* Virtual address of software ring */
+       uint64_t *ring_head_addr;  /* Virtual address of completion_head */
+       uint64_t shadow_completion_head; /* Shadow completion head value */
+       uint16_t head_free_desc;  /* Ring head */
+       uint16_t tail;  /* Ring tail */
+       /* Mask used to wrap enqueued descriptors on the sw ring */
+       uint32_t sw_ring_wrap_mask;
+       uint32_t irq_enable;  /* Enable ops dequeue interrupts if set to 1 */
+       uint8_t q_idx;  /* Queue index */
+       struct fpga_lte_fec_device *d;
+       /* MMIO register of shadow_tail used to enqueue descriptors */
+       void *shadow_tail_addr;
+};
+
+/* Write to 16 bit MMIO register address */
+static inline void
+mmio_write_16(void *addr, uint16_t value)
+{
+       *((volatile uint16_t *)(addr)) = rte_cpu_to_le_16(value);
+}
+
+/* Write to 32 bit MMIO register address */
+static inline void
+mmio_write_32(void *addr, uint32_t value)
+{
+       *((volatile uint32_t *)(addr)) = rte_cpu_to_le_32(value);
+}
+
+/* Write to 64 bit MMIO register address */
+static inline void
+mmio_write_64(void *addr, uint64_t value)
+{
+       *((volatile uint64_t *)(addr)) = rte_cpu_to_le_64(value);
+}
+
+/* Write a 8 bit register of a FPGA LTE FEC device */
+static inline void
+fpga_reg_write_8(void *mmio_base, uint32_t offset, uint8_t payload)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       *((volatile uint8_t *)(reg_addr)) = payload;
+}
+
+/* Write a 16 bit register of a FPGA LTE FEC device */
+static inline void
+fpga_reg_write_16(void *mmio_base, uint32_t offset, uint16_t payload)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       mmio_write_16(reg_addr, payload);
+}
+
+/* Write a 32 bit register of a FPGA LTE FEC device */
+static inline void
+fpga_reg_write_32(void *mmio_base, uint32_t offset, uint32_t payload)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       mmio_write_32(reg_addr, payload);
+}
+
+/* Write a 64 bit register of a FPGA LTE FEC device */
+static inline void
+fpga_reg_write_64(void *mmio_base, uint32_t offset, uint64_t payload)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       mmio_write_64(reg_addr, payload);
+}
+
+/* Write a ring control register of a FPGA LTE FEC device */
+static inline void
+fpga_ring_reg_write(void *mmio_base, uint32_t offset,
+               struct fpga_ring_ctrl_reg payload)
+{
+       fpga_reg_write_64(mmio_base, offset, payload.ring_base_addr);
+       fpga_reg_write_64(mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_ADDR,
+                       payload.ring_head_addr);
+       fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_SIZE,
+                       payload.ring_size);
+       fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_POINT,
+                       payload.head_point);
+       fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN,
+                       payload.flush_queue_en);
+       fpga_reg_write_16(mmio_base, offset + FPGA_LTE_FEC_RING_SHADOW_TAIL,
+                       payload.shadow_tail);
+       fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_MISC,
+                       payload.misc);
+       fpga_reg_write_8(mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
+                       payload.enable);
+}
+
+/* Read a register of FPGA LTE FEC device */
+static uint32_t
+fpga_reg_read_32(void *mmio_base, uint32_t offset)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       uint32_t ret = *((volatile uint32_t *)(reg_addr));
+       return rte_le_to_cpu_32(ret);
+}
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+/* Read a register of FPGA LTE FEC device */
+static uint8_t
+fpga_reg_read_8(void *mmio_base, uint32_t offset)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       return *((volatile uint8_t *)(reg_addr));
+}
+
+/* Read a register of FPGA LTE FEC device */
+static uint16_t
+fpga_reg_read_16(void *mmio_base, uint32_t offset)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       uint16_t ret = *((volatile uint16_t *)(reg_addr));
+       return rte_le_to_cpu_16(ret);
+}
+
+/* Read a register of FPGA LTE FEC device */
+static uint64_t
+fpga_reg_read_64(void *mmio_base, uint32_t offset)
+{
+       void *reg_addr = RTE_PTR_ADD(mmio_base, offset);
+       uint64_t ret = *((volatile uint64_t *)(reg_addr));
+       return rte_le_to_cpu_64(ret);
+}
+
+/* Read Ring Control Register of FPGA LTE FEC device */
+static inline void
+print_ring_reg_debug_info(void *mmio_base, uint32_t offset)
+{
+       rte_bbdev_log_debug(
+               "FPGA MMIO base address @ %p | Ring Control Register @ offset = 0x%08"
+               PRIx32, mmio_base, offset);
+       rte_bbdev_log_debug(
+               "RING_BASE_ADDR = 0x%016"PRIx64,
+               fpga_reg_read_64(mmio_base, offset));
+       rte_bbdev_log_debug(
+               "RING_HEAD_ADDR = 0x%016"PRIx64,
+               fpga_reg_read_64(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_HEAD_ADDR));
+       rte_bbdev_log_debug(
+               "RING_SIZE = 0x%04"PRIx16,
+               fpga_reg_read_16(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_SIZE));
+       rte_bbdev_log_debug(
+               "RING_MISC = 0x%02"PRIx8,
+               fpga_reg_read_8(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_MISC));
+       rte_bbdev_log_debug(
+               "RING_ENABLE = 0x%02"PRIx8,
+               fpga_reg_read_8(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_ENABLE));
+       rte_bbdev_log_debug(
+               "RING_FLUSH_QUEUE_EN = 0x%02"PRIx8,
+               fpga_reg_read_8(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN));
+       rte_bbdev_log_debug(
+               "RING_SHADOW_TAIL = 0x%04"PRIx16,
+               fpga_reg_read_16(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_SHADOW_TAIL));
+       rte_bbdev_log_debug(
+               "RING_HEAD_POINT = 0x%04"PRIx16,
+               fpga_reg_read_16(mmio_base, offset +
+                               FPGA_LTE_FEC_RING_HEAD_POINT));
+}
+
+/* Read Static Register of FPGA LTE FEC device */
+static inline void
+print_static_reg_debug_info(void *mmio_base)
+{
+       uint16_t config = fpga_reg_read_16(mmio_base,
+                       FPGA_LTE_FEC_CONFIGURATION);
+       uint8_t qmap_done = fpga_reg_read_8(mmio_base,
+                       FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE);
+       uint16_t lb_factor = fpga_reg_read_16(mmio_base,
+                       FPGA_LTE_FEC_LOAD_BALANCE_FACTOR);
+       uint16_t ring_desc_len = fpga_reg_read_16(mmio_base,
+                       FPGA_LTE_FEC_RING_DESC_LEN);
+       uint16_t flr_time_out = fpga_reg_read_16(mmio_base,
+                       FPGA_LTE_FEC_FLR_TIME_OUT);
+
+       rte_bbdev_log_debug("UL.DL Weights = %u.%u",
+                       ((uint8_t)config), ((uint8_t)(config >> 8)));
+       rte_bbdev_log_debug("UL.DL Load Balance = %u.%u",
+                       ((uint8_t)lb_factor), ((uint8_t)(lb_factor >> 8)));
+       rte_bbdev_log_debug("Queue-PF/VF Mapping Table = %s",
+                       (qmap_done > 0) ? "READY" : "NOT-READY");
+       rte_bbdev_log_debug("Ring Descriptor Size = %u bytes",
+                       ring_desc_len*FPGA_RING_DESC_LEN_UNIT_BYTES);
+       rte_bbdev_log_debug("FLR Timeout = %f usec",
+                       (float)flr_time_out*FPGA_FLR_TIMEOUT_UNIT);
+}
+
+/* Print decode DMA Descriptor of FPGA LTE FEC device */
+static void
+print_dma_dec_desc_debug_info(union fpga_dma_desc *desc)
+{
+       rte_bbdev_log_debug("DMA response desc %p\n"
+               "\t-- done(%"PRIu32") | iter(%"PRIu32") | crc_pass(%"PRIu32")"
+               " | error (%"PRIu32") | crc_type(%"PRIu32")\n"
+               "\t-- max_iter(%"PRIu32") | bypass_rm(%"PRIu32") | "
+               "irq_en (%"PRIu32") | drop_crc(%"PRIu32") | offset(%"PRIu32")\n"
+               "\t-- k(%"PRIu32") | in_len (%"PRIu16") | op_add(%p)\n"
+               "\t-- cbs_in_op(%"PRIu32") | in_add (0x%08"PRIx32"%08"PRIx32") | "
+               "out_add (0x%08"PRIx32"%08"PRIx32")",
+               desc,
+               (uint32_t)desc->dec_req.done,
+               (uint32_t)desc->dec_req.iter,
+               (uint32_t)desc->dec_req.crc_pass,
+               (uint32_t)desc->dec_req.error,
+               (uint32_t)desc->dec_req.crc_type,
+               (uint32_t)desc->dec_req.max_iter,
+               (uint32_t)desc->dec_req.bypass_rm,
+               (uint32_t)desc->dec_req.irq_en,
+               (uint32_t)desc->dec_req.drop_crc,
+               (uint32_t)desc->dec_req.offset,
+               (uint32_t)desc->dec_req.k,
+               (uint16_t)desc->dec_req.in_len,
+               desc->dec_req.op_addr,
+               (uint32_t)desc->dec_req.cbs_in_op,
+               (uint32_t)desc->dec_req.in_addr_hi,
+               (uint32_t)desc->dec_req.in_addr_lw,
+               (uint32_t)desc->dec_req.out_addr_hi,
+               (uint32_t)desc->dec_req.out_addr_lw);
+}
+#endif
+
+static int
+fpga_setup_queues(struct rte_bbdev *dev, uint16_t num_queues, int socket_id)
+{
+       /* Number of queues bound to a PF/VF */
+       uint32_t hw_q_num = 0;
+       uint32_t ring_size, payload, address, q_id, offset;
+       rte_iova_t phys_addr;
+       struct fpga_ring_ctrl_reg ring_reg;
+       struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
+
+       address = FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE;
+       if (!(fpga_reg_read_32(fpga_dev->mmio_base, address) & 0x1)) {
+               rte_bbdev_log(ERR,
+                               "Queue-PF/VF mapping is not set! Was PF configured for device (%s) ?",
+                               dev->data->name);
+               return -EPERM;
+       }
+
+       /* Clear queue registers structure */
+       memset(&ring_reg, 0, sizeof(struct fpga_ring_ctrl_reg));
+
+       /* Scan queue map.
+        * If a queue is valid and mapped to a calling PF/VF the read value is
+        * replaced with a queue ID and if it's not then
+        * FPGA_INVALID_HW_QUEUE_ID is returned.
+        */
+       for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
+               uint32_t hw_q_id = fpga_reg_read_32(fpga_dev->mmio_base,
+                               FPGA_LTE_FEC_QUEUE_MAP + (q_id << 2));
+
+               rte_bbdev_log_debug("%s: queue ID: %u, registry queue ID: %u",
+                               dev->device->name, q_id, hw_q_id);
+
+               if (hw_q_id != FPGA_INVALID_HW_QUEUE_ID) {
+                       fpga_dev->q_bound_bit_map |= (1ULL << q_id);
+                       /* Clear queue register of found queue */
+                       offset = FPGA_LTE_FEC_RING_CTRL_REGS +
+                               (sizeof(struct fpga_ring_ctrl_reg) * q_id);
+                       fpga_ring_reg_write(fpga_dev->mmio_base,
+                                       offset, ring_reg);
+                       ++hw_q_num;
+               }
+       }
+       if (hw_q_num == 0) {
+               rte_bbdev_log(ERR,
+                       "No HW queues assigned to this device. Probably this is a VF configured for PF mode. Check device configuration!");
+               return -ENODEV;
+       }
+
+       if (num_queues > hw_q_num) {
+               rte_bbdev_log(ERR,
+                       "Not enough queues for device %s! Requested: %u, available: %u",
+                       dev->device->name, num_queues, hw_q_num);
+               return -EINVAL;
+       }
+
+       ring_size = FPGA_RING_MAX_SIZE * sizeof(struct fpga_dma_dec_desc);
+
+       /* Enforce 32 byte alignment */
+       RTE_BUILD_BUG_ON((RTE_CACHE_LINE_SIZE % 32) != 0);
+
+       /* Allocate memory for SW descriptor rings */
+       fpga_dev->sw_rings = rte_zmalloc_socket(dev->device->driver->name,
+                       num_queues * ring_size, RTE_CACHE_LINE_SIZE,
+                       socket_id);
+       if (fpga_dev->sw_rings == NULL) {
+               rte_bbdev_log(ERR,
+                               "Failed to allocate memory for %s:%u sw_rings",
+                               dev->device->driver->name, dev->data->dev_id);
+               return -ENOMEM;
+       }
+
+       fpga_dev->sw_rings_phys = rte_malloc_virt2iova(fpga_dev->sw_rings);
+       fpga_dev->sw_ring_size = ring_size;
+       fpga_dev->sw_ring_max_depth = FPGA_RING_MAX_SIZE;
+
+       /* Allocate memory for ring flush status */
+       fpga_dev->flush_queue_status = rte_zmalloc_socket(NULL,
+                       sizeof(uint64_t), RTE_CACHE_LINE_SIZE, socket_id);
+       if (fpga_dev->flush_queue_status == NULL) {
+               rte_bbdev_log(ERR,
+                               "Failed to allocate memory for %s:%u flush_queue_status",
+                               dev->device->driver->name, dev->data->dev_id);
+               return -ENOMEM;
+       }
+
+       /* Set the flush status address registers */
+       phys_addr = rte_malloc_virt2iova(fpga_dev->flush_queue_status);
+
+       address = FPGA_LTE_FEC_VFQ_FLUSH_STATUS_LW;
+       payload = (uint32_t)(phys_addr);
+       fpga_reg_write_32(fpga_dev->mmio_base, address, payload);
+
+       address = FPGA_LTE_FEC_VFQ_FLUSH_STATUS_HI;
+       payload = (uint32_t)(phys_addr >> 32);
+       fpga_reg_write_32(fpga_dev->mmio_base, address, payload);
+
+       return 0;
+}
+
+static int
+fpga_dev_close(struct rte_bbdev *dev)
+{
+       struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
+
+       rte_free(fpga_dev->sw_rings);
+       rte_free(fpga_dev->flush_queue_status);
+
+       return 0;
+}
+
+static void
+fpga_dev_info_get(struct rte_bbdev *dev,
+               struct rte_bbdev_driver_info *dev_info)
+{
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+       uint32_t q_id = 0;
+
+       /* TODO RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN and numbers of buffers are set
+        * to temporary values as they are required by test application while
+        * validation phase.
+        */
+       static const struct rte_bbdev_op_cap bbdev_capabilities[] = {
+               {
+                       .type = RTE_BBDEV_OP_TURBO_DEC,
+                       .cap.turbo_dec = {
+                               .capability_flags =
+                                       RTE_BBDEV_TURBO_CRC_TYPE_24B |
+                                       RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE |
+                                       RTE_BBDEV_TURBO_DEC_INTERRUPTS |
+                                       RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN |
+                                       RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP,
+                               .max_llr_modulus = INT8_MAX,
+                               .num_buffers_src =
+                                               RTE_BBDEV_MAX_CODE_BLOCKS,
+                               .num_buffers_hard_out =
+                                       RTE_BBDEV_MAX_CODE_BLOCKS,
+                               .num_buffers_soft_out = 0
+                       }
+               },
+               {
+                       .type = RTE_BBDEV_OP_TURBO_ENC,
+                       .cap.turbo_enc = {
+                               .capability_flags =
+                                       RTE_BBDEV_TURBO_CRC_24B_ATTACH |
+                                       RTE_BBDEV_TURBO_RATE_MATCH |
+                                       RTE_BBDEV_TURBO_ENC_INTERRUPTS,
+                               .num_buffers_src =
+                                               RTE_BBDEV_MAX_CODE_BLOCKS,
+                               .num_buffers_dst =
+                                               RTE_BBDEV_MAX_CODE_BLOCKS
+                       }
+               },
+               RTE_BBDEV_END_OF_CAPABILITIES_LIST()
+       };
+
+       static struct rte_bbdev_queue_conf default_queue_conf;
+       default_queue_conf.socket = dev->data->socket_id;
+       default_queue_conf.queue_size = FPGA_RING_MAX_SIZE;
+
+
+       dev_info->driver_name = dev->device->driver->name;
+       dev_info->queue_size_lim = FPGA_RING_MAX_SIZE;
+       dev_info->hardware_accelerated = true;
+       dev_info->min_alignment = 64;
+       dev_info->default_queue_conf = default_queue_conf;
+       dev_info->capabilities = bbdev_capabilities;
+       dev_info->cpu_flag_reqs = NULL;
+
+       /* Calculates number of queues assigned to device */
+       dev_info->max_num_queues = 0;
+       for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
+               uint32_t hw_q_id = fpga_reg_read_32(d->mmio_base,
+                               FPGA_LTE_FEC_QUEUE_MAP + (q_id << 2));
+               if (hw_q_id != FPGA_INVALID_HW_QUEUE_ID)
+                       dev_info->max_num_queues++;
+       }
+}
+
+/**
+ * Find index of queue bound to current PF/VF which is unassigned. Return -1
+ * when there is no available queue
+ */
+static int
+fpga_find_free_queue_idx(struct rte_bbdev *dev,
+               const struct rte_bbdev_queue_conf *conf)
+{
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+       uint64_t q_idx;
+       uint8_t i = 0;
+       uint8_t range = FPGA_TOTAL_NUM_QUEUES >> 1;
+
+       if (conf->op_type == RTE_BBDEV_OP_TURBO_ENC) {
+               i = FPGA_NUM_DL_QUEUES;
+               range = FPGA_TOTAL_NUM_QUEUES;
+       }
+
+       for (; i < range; ++i) {
+               q_idx = 1ULL << i;
+               /* Check if index of queue is bound to current PF/VF */
+               if (d->q_bound_bit_map & q_idx)
+                       /* Check if found queue was not already assigned */
+                       if (!(d->q_assigned_bit_map & q_idx)) {
+                               d->q_assigned_bit_map |= q_idx;
+                               return i;
+                       }
+       }
+
+       rte_bbdev_log(INFO, "Failed to find free queue on %s", dev->data->name);
+
+       return -1;
+}
+
+static int
+fpga_queue_setup(struct rte_bbdev *dev, uint16_t queue_id,
+               const struct rte_bbdev_queue_conf *conf)
+{
+       uint32_t address, ring_offset;
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+       struct fpga_queue *q;
+       int8_t q_idx;
+
+       /* Check if there is a free queue to assign */
+       q_idx = fpga_find_free_queue_idx(dev, conf);
+       if (q_idx == -1)
+               return -1;
+
+       /* Allocate the queue data structure. */
+       q = rte_zmalloc_socket(dev->device->driver->name, sizeof(*q),
+                       RTE_CACHE_LINE_SIZE, conf->socket);
+       if (q == NULL) {
+               /* Mark queue as un-assigned */
+               d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
+               rte_bbdev_log(ERR, "Failed to allocate queue memory");
+               return -ENOMEM;
+       }
+
+       q->d = d;
+       q->q_idx = q_idx;
+
+       /* Set ring_base_addr */
+       q->ring_addr = RTE_PTR_ADD(d->sw_rings, (d->sw_ring_size * queue_id));
+       q->ring_ctrl_reg.ring_base_addr = d->sw_rings_phys +
+                       (d->sw_ring_size * queue_id);
+
+       /* Allocate memory for Completion Head variable*/
+       q->ring_head_addr = rte_zmalloc_socket(dev->device->driver->name,
+                       sizeof(uint64_t), RTE_CACHE_LINE_SIZE, conf->socket);
+       if (q->ring_head_addr == NULL) {
+               /* Mark queue as un-assigned */
+               d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
+               rte_free(q);
+               rte_bbdev_log(ERR,
+                               "Failed to allocate memory for %s:%u completion_head",
+                               dev->device->driver->name, dev->data->dev_id);
+               return -ENOMEM;
+       }
+       /* Set ring_head_addr */
+       q->ring_ctrl_reg.ring_head_addr =
+                       rte_malloc_virt2iova(q->ring_head_addr);
+
+       /* Clear shadow_completion_head */
+       q->shadow_completion_head = 0;
+
+       /* Set ring_size */
+       if (conf->queue_size > FPGA_RING_MAX_SIZE) {
+               /* Mark queue as un-assigned */
+               d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q_idx));
+               rte_free(q->ring_head_addr);
+               rte_free(q);
+               rte_bbdev_log(ERR,
+                               "Size of queue is too big %d (MAX: %d ) for %s:%u",
+                               conf->queue_size, FPGA_RING_MAX_SIZE,
+                               dev->device->driver->name, dev->data->dev_id);
+               return -EINVAL;
+       }
+       q->ring_ctrl_reg.ring_size = conf->queue_size;
+
+       /* Set Miscellaneous FPGA register*/
+       /* Max iteration number for TTI mitigation - todo */
+       q->ring_ctrl_reg.max_ul_dec = 0;
+       /* Enable max iteration number for TTI - todo */
+       q->ring_ctrl_reg.max_ul_dec_en = 0;
+
+       /* Enable the ring */
+       q->ring_ctrl_reg.enable = 1;
+
+       /* Set FPGA head_point and tail registers */
+       q->ring_ctrl_reg.head_point = q->tail = 0;
+
+       /* Set FPGA shadow_tail register */
+       q->ring_ctrl_reg.shadow_tail = q->tail;
+
+       /* Calculates the ring offset for found queue */
+       ring_offset = FPGA_LTE_FEC_RING_CTRL_REGS +
+                       (sizeof(struct fpga_ring_ctrl_reg) * q_idx);
+
+       /* Set FPGA Ring Control Registers */
+       fpga_ring_reg_write(d->mmio_base, ring_offset, q->ring_ctrl_reg);
+
+       /* Store MMIO register of shadow_tail */
+       address = ring_offset + FPGA_LTE_FEC_RING_SHADOW_TAIL;
+       q->shadow_tail_addr = RTE_PTR_ADD(d->mmio_base, address);
+
+       q->head_free_desc = q->tail;
+
+       /* Set wrap mask */
+       q->sw_ring_wrap_mask = conf->queue_size - 1;
+
+       rte_bbdev_log_debug("Setup dev%u q%u: queue_idx=%u",
+                       dev->data->dev_id, queue_id, q->q_idx);
+
+       dev->data->queues[queue_id].queue_private = q;
+
+       rte_bbdev_log_debug("BBDEV queue[%d] set up for FPGA queue[%d]",
+                       queue_id, q_idx);
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       /* Read FPGA Ring Control Registers after configuration*/
+       print_ring_reg_debug_info(d->mmio_base, ring_offset);
+#endif
+       return 0;
+}
+
+static int
+fpga_queue_release(struct rte_bbdev *dev, uint16_t queue_id)
+{
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+       struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
+       struct fpga_ring_ctrl_reg ring_reg;
+       uint32_t offset;
+
+       rte_bbdev_log_debug("FPGA Queue[%d] released", queue_id);
+
+       if (q != NULL) {
+               memset(&ring_reg, 0, sizeof(struct fpga_ring_ctrl_reg));
+               offset = FPGA_LTE_FEC_RING_CTRL_REGS +
+                       (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
+               /* Disable queue */
+               fpga_reg_write_8(d->mmio_base,
+                               offset + FPGA_LTE_FEC_RING_ENABLE, 0x00);
+               /* Clear queue registers */
+               fpga_ring_reg_write(d->mmio_base, offset, ring_reg);
+
+               /* Mark the Queue as un-assigned */
+               d->q_assigned_bit_map &= (0xFFFFFFFF - (1ULL << q->q_idx));
+               rte_free(q->ring_head_addr);
+               rte_free(q);
+               dev->data->queues[queue_id].queue_private = NULL;
+       }
+
+       return 0;
+}
+
+/* Function starts a device queue. */
+static int
+fpga_queue_start(struct rte_bbdev *dev, uint16_t queue_id)
+{
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       if (d == NULL) {
+               rte_bbdev_log(ERR, "Invalid device pointer");
+               return -1;
+       }
+#endif
+       struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
+       uint32_t offset = FPGA_LTE_FEC_RING_CTRL_REGS +
+                       (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
+       uint8_t enable = 0x01;
+       uint16_t zero = 0x0000;
+
+       /* Clear queue head and tail variables */
+       q->tail = q->head_free_desc = 0;
+
+       /* Clear FPGA head_point and tail registers */
+       fpga_reg_write_16(d->mmio_base, offset + FPGA_LTE_FEC_RING_HEAD_POINT,
+                       zero);
+       fpga_reg_write_16(d->mmio_base, offset + FPGA_LTE_FEC_RING_SHADOW_TAIL,
+                       zero);
+
+       /* Enable queue */
+       fpga_reg_write_8(d->mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
+                       enable);
+
+       rte_bbdev_log_debug("FPGA Queue[%d] started", queue_id);
+       return 0;
+}
+
+/* Function stops a device queue. */
+static int
+fpga_queue_stop(struct rte_bbdev *dev, uint16_t queue_id)
+{
+       struct fpga_lte_fec_device *d = dev->data->dev_private;
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       if (d == NULL) {
+               rte_bbdev_log(ERR, "Invalid device pointer");
+               return -1;
+       }
+#endif
+       struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
+       uint32_t offset = FPGA_LTE_FEC_RING_CTRL_REGS +
+                       (sizeof(struct fpga_ring_ctrl_reg) * q->q_idx);
+       uint8_t payload = 0x01;
+       uint8_t counter = 0;
+       uint8_t timeout = FPGA_QUEUE_FLUSH_TIMEOUT_US /
+                       FPGA_TIMEOUT_CHECK_INTERVAL;
+
+       /* Set flush_queue_en bit to trigger queue flushing */
+       fpga_reg_write_8(d->mmio_base,
+                       offset + FPGA_LTE_FEC_RING_FLUSH_QUEUE_EN, payload);
+
+       /** Check if queue flush is completed.
+        * FPGA will update the completion flag after queue flushing is
+        * completed. If completion flag is not updated within 1ms it is
+        * considered as a failure.
+        */
+       while (!(*((uint8_t *)d->flush_queue_status + q->q_idx) & payload)) {
+               if (counter > timeout) {
+                       rte_bbdev_log(ERR, "FPGA Queue Flush failed for queue %d",
+                                       queue_id);
+                       return -1;
+               }
+               usleep(FPGA_TIMEOUT_CHECK_INTERVAL);
+               counter++;
+       }
+
+       /* Disable queue */
+       payload = 0x00;
+       fpga_reg_write_8(d->mmio_base, offset + FPGA_LTE_FEC_RING_ENABLE,
+                       payload);
+
+       rte_bbdev_log_debug("FPGA Queue[%d] stopped", queue_id);
+       return 0;
+}
+
+static inline uint16_t
+get_queue_id(struct rte_bbdev_data *data, uint8_t q_idx)
+{
+       uint16_t queue_id;
+
+       for (queue_id = 0; queue_id < data->num_queues; ++queue_id) {
+               struct fpga_queue *q = data->queues[queue_id].queue_private;
+               if (q != NULL && q->q_idx == q_idx)
+                       return queue_id;
+       }
+
+       return -1;
+}
+
+/* Interrupt handler triggered by FPGA dev for handling specific interrupt */
+static void
+fpga_dev_interrupt_handler(void *cb_arg)
+{
+       struct rte_bbdev *dev = cb_arg;
+       struct fpga_lte_fec_device *fpga_dev = dev->data->dev_private;
+       struct fpga_queue *q;
+       uint64_t ring_head;
+       uint64_t q_idx;
+       uint16_t queue_id;
+       uint8_t i;
+
+       /* Scan queue assigned to this device */
+       for (i = 0; i < FPGA_TOTAL_NUM_QUEUES; ++i) {
+               q_idx = 1ULL << i;
+               if (fpga_dev->q_bound_bit_map & q_idx) {
+                       queue_id = get_queue_id(dev->data, i);
+                       if (queue_id == (uint16_t) -1)
+                               continue;
+
+                       /* Check if completion head was changed */
+                       q = dev->data->queues[queue_id].queue_private;
+                       ring_head = *q->ring_head_addr;
+                       if (q->shadow_completion_head != ring_head &&
+                               q->irq_enable == 1) {
+                               q->shadow_completion_head = ring_head;
+                               rte_bbdev_pmd_callback_process(
+                                               dev,
+                                               RTE_BBDEV_EVENT_DEQUEUE,
+                                               &queue_id);
+                       }
+               }
+       }
+}
+
+static int
+fpga_queue_intr_enable(struct rte_bbdev *dev, uint16_t queue_id)
+{
+       struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
+
+       if (!rte_intr_cap_multiple(dev->intr_handle))
+               return -ENOTSUP;
+
+       q->irq_enable = 1;
+
+       return 0;
+}
+
+static int
+fpga_queue_intr_disable(struct rte_bbdev *dev, uint16_t queue_id)
+{
+       struct fpga_queue *q = dev->data->queues[queue_id].queue_private;
+       q->irq_enable = 0;
+
+       return 0;
+}
+
+static int
+fpga_intr_enable(struct rte_bbdev *dev)
+{
+       int ret;
+       uint8_t i;
+
+       if (!rte_intr_cap_multiple(dev->intr_handle)) {
+               rte_bbdev_log(ERR, "Multiple intr vector is not supported by FPGA (%s)",
+                               dev->data->name);
+               return -ENOTSUP;
+       }
+
+       /* Create event file descriptors for each of 64 queue. Event fds will be
+        * mapped to FPGA IRQs in rte_intr_enable(). This is a 1:1 mapping where
+        * the IRQ number is a direct translation to the queue number.
+        *
+        * 63 (FPGA_NUM_INTR_VEC) event fds are created as rte_intr_enable()
+        * mapped the first IRQ to already created interrupt event file
+        * descriptor (intr_handle->fd).
+        */
+       if (rte_intr_efd_enable(dev->intr_handle, FPGA_NUM_INTR_VEC)) {
+               rte_bbdev_log(ERR, "Failed to create fds for %u queues",
+                               dev->data->num_queues);
+               return -1;
+       }
+
+       /* TODO Each event file descriptor is overwritten by interrupt event
+        * file descriptor. That descriptor is added to epoll observed list.
+        * It ensures that callback function assigned to that descriptor will
+        * invoked when any FPGA queue issues interrupt.
+        */
+       for (i = 0; i < FPGA_NUM_INTR_VEC; ++i)
+               dev->intr_handle->efds[i] = dev->intr_handle->fd;
+
+       if (!dev->intr_handle->intr_vec) {
+               dev->intr_handle->intr_vec = rte_zmalloc("intr_vec",
+                               dev->data->num_queues * sizeof(int), 0);
+               if (!dev->intr_handle->intr_vec) {
+                       rte_bbdev_log(ERR, "Failed to allocate %u vectors",
+                                       dev->data->num_queues);
+                       return -ENOMEM;
+               }
+       }
+
+       ret = rte_intr_enable(dev->intr_handle);
+       if (ret < 0) {
+               rte_bbdev_log(ERR,
+                               "Couldn't enable interrupts for device: %s",
+                               dev->data->name);
+               return ret;
+       }
+
+       ret = rte_intr_callback_register(dev->intr_handle,
+                       fpga_dev_interrupt_handler, dev);
+       if (ret < 0) {
+               rte_bbdev_log(ERR,
+                               "Couldn't register interrupt callback for device: %s",
+                               dev->data->name);
+               return ret;
+       }
+
+       return 0;
+}
+
+static const struct rte_bbdev_ops fpga_ops = {
+       .setup_queues = fpga_setup_queues,
+       .intr_enable = fpga_intr_enable,
+       .close = fpga_dev_close,
+       .info_get = fpga_dev_info_get,
+       .queue_setup = fpga_queue_setup,
+       .queue_stop = fpga_queue_stop,
+       .queue_start = fpga_queue_start,
+       .queue_release = fpga_queue_release,
+       .queue_intr_enable = fpga_queue_intr_enable,
+       .queue_intr_disable = fpga_queue_intr_disable
+};
+
+static inline void
+fpga_dma_enqueue(struct fpga_queue *q, uint16_t num_desc,
+               struct rte_bbdev_stats *queue_stats)
+{
+#ifdef RTE_BBDEV_OFFLOAD_COST
+       uint64_t start_time = 0;
+       queue_stats->acc_offload_cycles = 0;
+#else
+       RTE_SET_USED(queue_stats);
+#endif
+
+       /* Update tail and shadow_tail register */
+       q->tail = (q->tail + num_desc) & q->sw_ring_wrap_mask;
+
+       rte_wmb();
+
+#ifdef RTE_BBDEV_OFFLOAD_COST
+       /* Start time measurement for enqueue function offload. */
+       start_time = rte_rdtsc_precise();
+#endif
+       mmio_write_16(q->shadow_tail_addr, q->tail);
+
+#ifdef RTE_BBDEV_OFFLOAD_COST
+       rte_wmb();
+       queue_stats->acc_offload_cycles += rte_rdtsc_precise() - start_time;
+#endif
+}
+
+/* Calculates number of CBs in processed encoder TB based on 'r' and input
+ * length.
+ */
+static inline uint8_t
+get_num_cbs_in_op_enc(struct rte_bbdev_op_turbo_enc *turbo_enc)
+{
+       uint8_t c, c_neg, r, crc24_bits = 0;
+       uint16_t k, k_neg, k_pos;
+       uint8_t cbs_in_op = 0;
+       int32_t length;
+
+       length = turbo_enc->input.length;
+       r = turbo_enc->tb_params.r;
+       c = turbo_enc->tb_params.c;
+       c_neg = turbo_enc->tb_params.c_neg;
+       k_neg = turbo_enc->tb_params.k_neg;
+       k_pos = turbo_enc->tb_params.k_pos;
+       crc24_bits = 24;
+       while (length > 0 && r < c) {
+               k = (r < c_neg) ? k_neg : k_pos;
+               length -= (k - crc24_bits) >> 3;
+               r++;
+               cbs_in_op++;
+       }
+
+       return cbs_in_op;
+}
+
+/* Calculates number of CBs in processed decoder TB based on 'r' and input
+ * length.
+ */
+static inline uint16_t
+get_num_cbs_in_op_dec(struct rte_bbdev_op_turbo_dec *turbo_dec)
+{
+       uint8_t c, c_neg, r = 0;
+       uint16_t kw, k, k_neg, k_pos, cbs_in_op = 0;
+       int32_t length;
+
+       length = turbo_dec->input.length;
+       r = turbo_dec->tb_params.r;
+       c = turbo_dec->tb_params.c;
+       c_neg = turbo_dec->tb_params.c_neg;
+       k_neg = turbo_dec->tb_params.k_neg;
+       k_pos = turbo_dec->tb_params.k_pos;
+       while (length > 0 && r < c) {
+               k = (r < c_neg) ? k_neg : k_pos;
+               kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
+               length -= kw;
+               r++;
+               cbs_in_op++;
+       }
+
+       return cbs_in_op;
+}
+
+/* Read flag value 0/1/ from bitmap */
+static inline bool
+check_bit(uint32_t bitmap, uint32_t bitmask)
+{
+       return bitmap & bitmask;
+}
+
+/* Print an error if a descriptor error has occurred.
+ *  Return 0 on success, 1 on failure
+ */
+static inline int
+check_desc_error(uint32_t error_code) {
+       switch (error_code) {
+       case DESC_ERR_NO_ERR:
+               return 0;
+       case DESC_ERR_K_OUT_OF_RANGE:
+               rte_bbdev_log(ERR, "Block_size_k is out of range (k<40 or k>6144)");
+               break;
+       case DESC_ERR_K_NOT_NORMAL:
+               rte_bbdev_log(ERR, "Block_size_k is not a normal value within normal range");
+               break;
+       case DESC_ERR_KPAI_NOT_NORMAL:
+               rte_bbdev_log(ERR, "Three_kpai is not a normal value for UL only");
+               break;
+       case DESC_ERR_DESC_OFFSET_ERR:
+               rte_bbdev_log(ERR, "Queue offset does not meet the expectation in the FPGA");
+               break;
+       case (DESC_ERR_K_OUT_OF_RANGE | DESC_ERR_DESC_OFFSET_ERR):
+               rte_bbdev_log(ERR, "Block_size_k is out of range (k<40 or k>6144) and queue offset error");
+               break;
+       case (DESC_ERR_K_NOT_NORMAL | DESC_ERR_DESC_OFFSET_ERR):
+               rte_bbdev_log(ERR, "Block_size_k is not a normal value within normal range and queue offset error");
+               break;
+       case (DESC_ERR_KPAI_NOT_NORMAL | DESC_ERR_DESC_OFFSET_ERR):
+               rte_bbdev_log(ERR, "Three_kpai is not a normal value for UL only and queue offset error");
+               break;
+       case DESC_ERR_DESC_READ_FAIL:
+               rte_bbdev_log(ERR, "Unsuccessful completion for descriptor read");
+               break;
+       case DESC_ERR_DESC_READ_TIMEOUT:
+               rte_bbdev_log(ERR, "Descriptor read time-out");
+               break;
+       case DESC_ERR_DESC_READ_TLP_POISONED:
+               rte_bbdev_log(ERR, "Descriptor read TLP poisoned");
+               break;
+       case DESC_ERR_CB_READ_FAIL:
+               rte_bbdev_log(ERR, "Unsuccessful completion for code block");
+               break;
+       case DESC_ERR_CB_READ_TIMEOUT:
+               rte_bbdev_log(ERR, "Code block read time-out");
+               break;
+       case DESC_ERR_CB_READ_TLP_POISONED:
+               rte_bbdev_log(ERR, "Code block read TLP poisoned");
+               break;
+       default:
+               rte_bbdev_log(ERR, "Descriptor error unknown error code %u",
+                               error_code);
+               break;
+       }
+       return 1;
+}
+
+/**
+ * Set DMA descriptor for encode operation (1 Code Block)
+ *
+ * @param op
+ *   Pointer to a single encode operation.
+ * @param desc
+ *   Pointer to DMA descriptor.
+ * @param input
+ *   Pointer to pointer to input data which will be decoded.
+ * @param k
+ *   K value (length of input in bits).
+ * @param e
+ *   E value (length of output in bits).
+ * @param ncb
+ *   Ncb value (size of the soft buffer).
+ * @param out_length
+ *   Length of output buffer
+ * @param in_offset
+ *   Input offset in rte_mbuf structure. It is used for calculating the point
+ *   where data is starting.
+ * @param out_offset
+ *   Output offset in rte_mbuf structure. It is used for calculating the point
+ *   where hard output data will be stored.
+ * @param cbs_in_op
+ *   Number of CBs contained in one operation.
+ */
+static inline int
+fpga_dma_desc_te_fill(struct rte_bbdev_enc_op *op,
+               struct fpga_dma_enc_desc *desc, struct rte_mbuf *input,
+               struct rte_mbuf *output, uint16_t k, uint16_t e, uint16_t ncb,
+               uint32_t in_offset, uint32_t out_offset, uint16_t desc_offset,
+               uint8_t cbs_in_op)
+
+{
+       /* reset */
+       desc->done = 0;
+       desc->crc_en = check_bit(op->turbo_enc.op_flags,
+               RTE_BBDEV_TURBO_CRC_24B_ATTACH);
+       desc->bypass_rm = !check_bit(op->turbo_enc.op_flags,
+               RTE_BBDEV_TURBO_RATE_MATCH);
+       desc->k = k;
+       desc->e = e;
+       desc->ncb = ncb;
+       desc->rv = op->turbo_enc.rv_index;
+       desc->offset = desc_offset;
+       /* Set inbound data buffer address */
+       desc->in_addr_hi = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(input, in_offset) >> 32);
+       desc->in_addr_lw = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(input, in_offset));
+
+       desc->out_addr_hi = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(output, out_offset) >> 32);
+       desc->out_addr_lw = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(output, out_offset));
+
+       /* Save software context needed for dequeue */
+       desc->op_addr = op;
+
+       /* Set total number of CBs in an op */
+       desc->cbs_in_op = cbs_in_op;
+
+       return 0;
+}
+
+/**
+ * Set DMA descriptor for encode operation (1 Code Block)
+ *
+ * @param op
+ *   Pointer to a single encode operation.
+ * @param desc
+ *   Pointer to DMA descriptor.
+ * @param input
+ *   Pointer to pointer to input data which will be decoded.
+ * @param in_length
+ *   Length of an input.
+ * @param k
+ *   K value (length of an output in bits).
+ * @param in_offset
+ *   Input offset in rte_mbuf structure. It is used for calculating the point
+ *   where data is starting.
+ * @param out_offset
+ *   Output offset in rte_mbuf structure. It is used for calculating the point
+ *   where hard output data will be stored.
+ * @param cbs_in_op
+ *   Number of CBs contained in one operation.
+ */
+static inline int
+fpga_dma_desc_td_fill(struct rte_bbdev_dec_op *op,
+               struct fpga_dma_dec_desc *desc, struct rte_mbuf *input,
+               struct rte_mbuf *output, uint16_t in_length, uint16_t k,
+               uint32_t in_offset, uint32_t out_offset, uint16_t desc_offset,
+               uint8_t cbs_in_op)
+{
+       /* reset */
+       desc->done = 0;
+       /* Set inbound data buffer address */
+       desc->in_addr_hi = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(input, in_offset) >> 32);
+       desc->in_addr_lw = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(input, in_offset));
+       desc->in_len = in_length;
+       desc->k = k;
+       desc->crc_type = !check_bit(op->turbo_dec.op_flags,
+                       RTE_BBDEV_TURBO_CRC_TYPE_24B);
+       if ((op->turbo_dec.code_block_mode == 0)
+               && !check_bit(op->turbo_dec.op_flags,
+               RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP))
+               desc->drop_crc = 1;
+       desc->max_iter = op->turbo_dec.iter_max * 2;
+       desc->offset = desc_offset;
+       desc->out_addr_hi = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(output, out_offset) >> 32);
+       desc->out_addr_lw = (uint32_t)(
+                       rte_pktmbuf_mtophys_offset(output, out_offset));
+
+       /* Save software context needed for dequeue */
+       desc->op_addr = op;
+
+       /* Set total number of CBs in an op */
+       desc->cbs_in_op = cbs_in_op;
+
+       return 0;
+}
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+/* Validates turbo encoder parameters */
+static int
+validate_enc_op(struct rte_bbdev_enc_op *op)
+{
+       struct rte_bbdev_op_turbo_enc *turbo_enc = &op->turbo_enc;
+       struct rte_bbdev_op_enc_cb_params *cb = NULL;
+       struct rte_bbdev_op_enc_tb_params *tb = NULL;
+       uint16_t kw, kw_neg, kw_pos;
+
+       if (turbo_enc->input.length >
+                       RTE_BBDEV_MAX_TB_SIZE >> 3) {
+               rte_bbdev_log(ERR, "TB size (%u) is too big, max: %d",
+                               turbo_enc->input.length, RTE_BBDEV_MAX_TB_SIZE);
+               op->status = 1 << RTE_BBDEV_DATA_ERROR;
+               return -1;
+       }
+
+       if (op->mempool == NULL) {
+               rte_bbdev_log(ERR, "Invalid mempool pointer");
+               return -1;
+       }
+       if (turbo_enc->input.data == NULL) {
+               rte_bbdev_log(ERR, "Invalid input pointer");
+               return -1;
+       }
+       if (turbo_enc->output.data == NULL) {
+               rte_bbdev_log(ERR, "Invalid output pointer");
+               return -1;
+       }
+       if (turbo_enc->rv_index > 3) {
+               rte_bbdev_log(ERR,
+                               "rv_index (%u) is out of range 0 <= value <= 3",
+                               turbo_enc->rv_index);
+               return -1;
+       }
+       if (turbo_enc->code_block_mode != 0 &&
+                       turbo_enc->code_block_mode != 1) {
+               rte_bbdev_log(ERR,
+                               "code_block_mode (%u) is out of range 0 <= value <= 1",
+                               turbo_enc->code_block_mode);
+               return -1;
+       }
+
+       if (turbo_enc->code_block_mode == 0) {
+               tb = &turbo_enc->tb_params;
+               if ((tb->k_neg < RTE_BBDEV_MIN_CB_SIZE
+                               || tb->k_neg > RTE_BBDEV_MAX_CB_SIZE)
+                               && tb->c_neg > 0) {
+                       rte_bbdev_log(ERR,
+                                       "k_neg (%u) is out of range %u <= value <= %u",
+                                       tb->k_neg, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+               if (tb->k_pos < RTE_BBDEV_MIN_CB_SIZE
+                               || tb->k_pos > RTE_BBDEV_MAX_CB_SIZE) {
+                       rte_bbdev_log(ERR,
+                                       "k_pos (%u) is out of range %u <= value <= %u",
+                                       tb->k_pos, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+               if (tb->c_neg > (RTE_BBDEV_MAX_CODE_BLOCKS - 1))
+                       rte_bbdev_log(ERR,
+                                       "c_neg (%u) is out of range 0 <= value <= %u",
+                                       tb->c_neg,
+                                       RTE_BBDEV_MAX_CODE_BLOCKS - 1);
+               if (tb->c < 1 || tb->c > RTE_BBDEV_MAX_CODE_BLOCKS) {
+                       rte_bbdev_log(ERR,
+                                       "c (%u) is out of range 1 <= value <= %u",
+                                       tb->c, RTE_BBDEV_MAX_CODE_BLOCKS);
+                       return -1;
+               }
+               if (tb->cab > tb->c) {
+                       rte_bbdev_log(ERR,
+                                       "cab (%u) is greater than c (%u)",
+                                       tb->cab, tb->c);
+                       return -1;
+               }
+               if ((tb->ea < RTE_BBDEV_MIN_CB_SIZE || (tb->ea % 2))
+                               && tb->r < tb->cab) {
+                       rte_bbdev_log(ERR,
+                                       "ea (%u) is less than %u or it is not even",
+                                       tb->ea, RTE_BBDEV_MIN_CB_SIZE);
+                       return -1;
+               }
+               if ((tb->eb < RTE_BBDEV_MIN_CB_SIZE || (tb->eb % 2))
+                               && tb->c > tb->cab) {
+                       rte_bbdev_log(ERR,
+                                       "eb (%u) is less than %u or it is not even",
+                                       tb->eb, RTE_BBDEV_MIN_CB_SIZE);
+                       return -1;
+               }
+
+               kw_neg = 3 * RTE_ALIGN_CEIL(tb->k_neg + 4,
+                                       RTE_BBDEV_C_SUBBLOCK);
+               if (tb->ncb_neg < tb->k_neg || tb->ncb_neg > kw_neg) {
+                       rte_bbdev_log(ERR,
+                                       "ncb_neg (%u) is out of range (%u) k_neg <= value <= (%u) kw_neg",
+                                       tb->ncb_neg, tb->k_neg, kw_neg);
+                       return -1;
+               }
+
+               kw_pos = 3 * RTE_ALIGN_CEIL(tb->k_pos + 4,
+                                       RTE_BBDEV_C_SUBBLOCK);
+               if (tb->ncb_pos < tb->k_pos || tb->ncb_pos > kw_pos) {
+                       rte_bbdev_log(ERR,
+                                       "ncb_pos (%u) is out of range (%u) k_pos <= value <= (%u) kw_pos",
+                                       tb->ncb_pos, tb->k_pos, kw_pos);
+                       return -1;
+               }
+               if (tb->r > (tb->c - 1)) {
+                       rte_bbdev_log(ERR,
+                                       "r (%u) is greater than c - 1 (%u)",
+                                       tb->r, tb->c - 1);
+                       return -1;
+               }
+       } else {
+               cb = &turbo_enc->cb_params;
+               if (cb->k < RTE_BBDEV_MIN_CB_SIZE
+                               || cb->k > RTE_BBDEV_MAX_CB_SIZE) {
+                       rte_bbdev_log(ERR,
+                                       "k (%u) is out of range %u <= value <= %u",
+                                       cb->k, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+
+               if (cb->e < RTE_BBDEV_MIN_CB_SIZE || (cb->e % 2)) {
+                       rte_bbdev_log(ERR,
+                                       "e (%u) is less than %u or it is not even",
+                                       cb->e, RTE_BBDEV_MIN_CB_SIZE);
+                       return -1;
+               }
+
+               kw = RTE_ALIGN_CEIL(cb->k + 4, RTE_BBDEV_C_SUBBLOCK) * 3;
+               if (cb->ncb < cb->k || cb->ncb > kw) {
+                       rte_bbdev_log(ERR,
+                                       "ncb (%u) is out of range (%u) k <= value <= (%u) kw",
+                                       cb->ncb, cb->k, kw);
+                       return -1;
+               }
+       }
+
+       return 0;
+}
+#endif
+
+static inline char *
+mbuf_append(struct rte_mbuf *m_head, struct rte_mbuf *m, uint16_t len)
+{
+       if (unlikely(len > rte_pktmbuf_tailroom(m)))
+               return NULL;
+
+       char *tail = (char *)m->buf_addr + m->data_off + m->data_len;
+       m->data_len = (uint16_t)(m->data_len + len);
+       m_head->pkt_len  = (m_head->pkt_len + len);
+       return tail;
+}
+
+static inline int
+enqueue_enc_one_op_cb(struct fpga_queue *q, struct rte_bbdev_enc_op *op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       struct rte_mbuf *input;
+       struct rte_mbuf *output;
+       int ret;
+       uint16_t k, e, ncb, ring_offset;
+       uint32_t total_left, in_length, out_length, in_offset, out_offset;
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       /* Validate op structure */
+       if (validate_enc_op(op) == -1) {
+               rte_bbdev_log(ERR, "Turbo encoder validation failed");
+               return -EINVAL;
+       }
+#endif
+
+       input = op->turbo_enc.input.data;
+       output = op->turbo_enc.output.data;
+       in_offset = op->turbo_enc.input.offset;
+       out_offset = op->turbo_enc.output.offset;
+       total_left = op->turbo_enc.input.length;
+       k = op->turbo_enc.cb_params.k;
+       e = op->turbo_enc.cb_params.e;
+       ncb = op->turbo_enc.cb_params.ncb;
+
+       if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_CRC_24B_ATTACH))
+               in_length = ((k - 24) >> 3);
+       else
+               in_length = k >> 3;
+
+       if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_RATE_MATCH))
+               out_length = (e + 7) >> 3;
+       else
+               out_length = (k >> 3) * 3 + 2;
+
+       mbuf_append(output, output, out_length);
+
+       /* Offset into the ring */
+       ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
+       /* Setup DMA Descriptor */
+       desc = q->ring_addr + ring_offset;
+
+       ret = fpga_dma_desc_te_fill(op, &desc->enc_req, input, output, k, e,
+                       ncb, in_offset, out_offset, ring_offset, 1);
+       if (unlikely(ret < 0))
+               return ret;
+
+       /* Update lengths */
+       total_left -= in_length;
+       op->turbo_enc.output.length += out_length;
+
+       if (total_left > 0) {
+               rte_bbdev_log(ERR,
+                       "Mismatch between mbuf length and included CB sizes: mbuf len %u, cb len %u",
+                               total_left, in_length);
+               return -1;
+       }
+
+       return 1;
+}
+
+static inline int
+enqueue_enc_one_op_tb(struct fpga_queue *q, struct rte_bbdev_enc_op *op,
+               uint16_t desc_offset, uint8_t cbs_in_op)
+{
+       union fpga_dma_desc *desc;
+       struct rte_mbuf *input, *output_head, *output;
+       int ret;
+       uint8_t r, c, crc24_bits = 0;
+       uint16_t k, e, ncb, ring_offset;
+       uint32_t mbuf_total_left, in_length, out_length, in_offset, out_offset;
+       uint32_t seg_total_left;
+       uint16_t current_enqueued_cbs = 0;
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       /* Validate op structure */
+       if (validate_enc_op(op) == -1) {
+               rte_bbdev_log(ERR, "Turbo encoder validation failed");
+               return -EINVAL;
+       }
+#endif
+
+       input = op->turbo_enc.input.data;
+       output_head = output = op->turbo_enc.output.data;
+       in_offset = op->turbo_enc.input.offset;
+       out_offset = op->turbo_enc.output.offset;
+       mbuf_total_left = op->turbo_enc.input.length;
+
+       c = op->turbo_enc.tb_params.c;
+       r = op->turbo_enc.tb_params.r;
+
+       if (check_bit(op->turbo_enc.op_flags, RTE_BBDEV_TURBO_CRC_24B_ATTACH))
+               crc24_bits = 24;
+
+       while (mbuf_total_left > 0 && r < c && input != NULL) {
+               seg_total_left = rte_pktmbuf_data_len(input) - in_offset;
+
+               e = (r < op->turbo_enc.tb_params.cab) ?
+                               op->turbo_enc.tb_params.ea :
+                               op->turbo_enc.tb_params.eb;
+               k = (r < op->turbo_enc.tb_params.c_neg) ?
+                               op->turbo_enc.tb_params.k_neg :
+                               op->turbo_enc.tb_params.k_pos;
+               ncb = (r < op->turbo_enc.tb_params.c_neg) ?
+                               op->turbo_enc.tb_params.ncb_neg :
+                               op->turbo_enc.tb_params.ncb_pos;
+
+               in_length = ((k - crc24_bits) >> 3);
+
+               if (check_bit(op->turbo_enc.op_flags,
+                       RTE_BBDEV_TURBO_RATE_MATCH))
+                       out_length = (e + 7) >> 3;
+               else
+                       out_length = (k >> 3) * 3 + 2;
+
+               mbuf_append(output_head, output, out_length);
+
+               /* Setup DMA Descriptor */
+               ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
+               desc = q->ring_addr + ring_offset;
+               ret = fpga_dma_desc_te_fill(op, &desc->enc_req, input, output,
+                               k, e, ncb, in_offset, out_offset, ring_offset,
+                               cbs_in_op);
+               if (unlikely(ret < 0))
+                       return ret;
+
+               rte_bbdev_log_debug("DMA request desc %p", desc);
+
+               /* Update lengths */
+               op->turbo_enc.output.length += out_length;
+               mbuf_total_left -= in_length;
+
+               /* Update offsets */
+               if (seg_total_left == in_length) {
+                       /* Go to the next mbuf */
+                       input = input->next;
+                       output = output->next;
+                       in_offset = 0;
+                       out_offset = 0;
+               } else {
+                       in_offset += in_length;
+                       out_offset += out_length;
+               }
+
+               r++;
+               desc_offset++;
+               current_enqueued_cbs++;
+       }
+
+       if (mbuf_total_left > 0) {
+               rte_bbdev_log(ERR,
+                               "Some date still left for processing: mbuf_total_left = %u",
+                               mbuf_total_left);
+               return -1;
+       }
+
+       return current_enqueued_cbs;
+}
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+/* Validates turbo decoder parameters */
+static int
+validate_dec_op(struct rte_bbdev_dec_op *op)
+{
+       struct rte_bbdev_op_turbo_dec *turbo_dec = &op->turbo_dec;
+       struct rte_bbdev_op_dec_cb_params *cb = NULL;
+       struct rte_bbdev_op_dec_tb_params *tb = NULL;
+
+       if (op->mempool == NULL) {
+               rte_bbdev_log(ERR, "Invalid mempool pointer");
+               return -1;
+       }
+       if (turbo_dec->input.data == NULL) {
+               rte_bbdev_log(ERR, "Invalid input pointer");
+               return -1;
+       }
+       if (turbo_dec->hard_output.data == NULL) {
+               rte_bbdev_log(ERR, "Invalid hard_output pointer");
+               return -1;
+       }
+       if (turbo_dec->rv_index > 3) {
+               rte_bbdev_log(ERR,
+                               "rv_index (%u) is out of range 0 <= value <= 3",
+                               turbo_dec->rv_index);
+               return -1;
+       }
+       if (turbo_dec->iter_min < 1) {
+               rte_bbdev_log(ERR,
+                               "iter_min (%u) is less than 1",
+                               turbo_dec->iter_min);
+               return -1;
+       }
+       if (turbo_dec->iter_max <= 2) {
+               rte_bbdev_log(ERR,
+                               "iter_max (%u) is less than or equal to 2",
+                               turbo_dec->iter_max);
+               return -1;
+       }
+       if (turbo_dec->iter_min > turbo_dec->iter_max) {
+               rte_bbdev_log(ERR,
+                               "iter_min (%u) is greater than iter_max (%u)",
+                               turbo_dec->iter_min, turbo_dec->iter_max);
+               return -1;
+       }
+       if (turbo_dec->code_block_mode != 0 &&
+                       turbo_dec->code_block_mode != 1) {
+               rte_bbdev_log(ERR,
+                               "code_block_mode (%u) is out of range 0 <= value <= 1",
+                               turbo_dec->code_block_mode);
+               return -1;
+       }
+
+       if (turbo_dec->code_block_mode == 0) {
+
+               if ((turbo_dec->op_flags &
+                       RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP) &&
+                       !(turbo_dec->op_flags & RTE_BBDEV_TURBO_CRC_TYPE_24B)) {
+                       rte_bbdev_log(ERR,
+                               "RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP should accompany RTE_BBDEV_TURBO_CRC_TYPE_24B");
+                       return -1;
+               }
+
+               tb = &turbo_dec->tb_params;
+               if ((tb->k_neg < RTE_BBDEV_MIN_CB_SIZE
+                               || tb->k_neg > RTE_BBDEV_MAX_CB_SIZE)
+                               && tb->c_neg > 0) {
+                       rte_bbdev_log(ERR,
+                                       "k_neg (%u) is out of range %u <= value <= %u",
+                                       tb->k_neg, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+               if ((tb->k_pos < RTE_BBDEV_MIN_CB_SIZE
+                               || tb->k_pos > RTE_BBDEV_MAX_CB_SIZE)
+                               && tb->c > tb->c_neg) {
+                       rte_bbdev_log(ERR,
+                                       "k_pos (%u) is out of range %u <= value <= %u",
+                                       tb->k_pos, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+               if (tb->c_neg > (RTE_BBDEV_MAX_CODE_BLOCKS - 1))
+                       rte_bbdev_log(ERR,
+                                       "c_neg (%u) is out of range 0 <= value <= %u",
+                                       tb->c_neg,
+                                       RTE_BBDEV_MAX_CODE_BLOCKS - 1);
+               if (tb->c < 1 || tb->c > RTE_BBDEV_MAX_CODE_BLOCKS) {
+                       rte_bbdev_log(ERR,
+                                       "c (%u) is out of range 1 <= value <= %u",
+                                       tb->c, RTE_BBDEV_MAX_CODE_BLOCKS);
+                       return -1;
+               }
+               if (tb->cab > tb->c) {
+                       rte_bbdev_log(ERR,
+                                       "cab (%u) is greater than c (%u)",
+                                       tb->cab, tb->c);
+                       return -1;
+               }
+       } else {
+
+               if (turbo_dec->op_flags & RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP) {
+                       rte_bbdev_log(ERR,
+                                       "RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP is invalid in CB-mode");
+                       return -1;
+               }
+
+               cb = &turbo_dec->cb_params;
+               if (cb->k < RTE_BBDEV_MIN_CB_SIZE
+                               || cb->k > RTE_BBDEV_MAX_CB_SIZE) {
+                       rte_bbdev_log(ERR,
+                                       "k (%u) is out of range %u <= value <= %u",
+                                       cb->k, RTE_BBDEV_MIN_CB_SIZE,
+                                       RTE_BBDEV_MAX_CB_SIZE);
+                       return -1;
+               }
+       }
+
+       return 0;
+}
+#endif
+
+static inline int
+enqueue_dec_one_op_cb(struct fpga_queue *q, struct rte_bbdev_dec_op *op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       struct rte_mbuf *input;
+       struct rte_mbuf *output;
+       int ret;
+       uint16_t k, kw, ring_offset;
+       uint32_t total_left, in_length, out_length, in_offset, out_offset;
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       /* Validate op structure */
+       if (validate_dec_op(op) == -1) {
+               rte_bbdev_log(ERR, "Turbo decoder validation failed");
+               return -EINVAL;
+       }
+#endif
+
+       input = op->turbo_dec.input.data;
+       output = op->turbo_dec.hard_output.data;
+       total_left = op->turbo_dec.input.length;
+       in_offset = op->turbo_dec.input.offset;
+       out_offset = op->turbo_dec.hard_output.offset;
+
+       k = op->turbo_dec.cb_params.k;
+       kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
+       in_length = kw;
+       out_length = k >> 3;
+
+       mbuf_append(output, output, out_length);
+
+       /* Setup DMA Descriptor */
+       ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
+       desc = q->ring_addr + ring_offset;
+       ret = fpga_dma_desc_td_fill(op, &desc->dec_req, input, output,
+                       in_length, k, in_offset, out_offset, ring_offset, 1);
+       if (unlikely(ret < 0))
+               return ret;
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       print_dma_dec_desc_debug_info(desc);
+#endif
+
+       /* Update lengths */
+       total_left -= in_length;
+       op->turbo_dec.hard_output.length += out_length;
+
+       if (total_left > 0) {
+               rte_bbdev_log(ERR,
+                               "Mismatch between mbuf length and included CB sizes: mbuf len %u, cb len %u",
+                               total_left, in_length);
+               return -1;
+       }
+
+       return 1;
+}
+
+
+static inline int
+enqueue_dec_one_op_tb(struct fpga_queue *q, struct rte_bbdev_dec_op *op,
+               uint16_t desc_offset, uint8_t cbs_in_op)
+{
+       union fpga_dma_desc *desc;
+       struct rte_mbuf *input, *output_head, *output;
+       int ret;
+       uint8_t r, c;
+       uint16_t k, kw, in_length, out_length, ring_offset;
+       uint32_t mbuf_total_left, seg_total_left, in_offset, out_offset;
+       uint16_t current_enqueued_cbs = 0;
+       uint16_t crc24_overlap = 0;
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       /* Validate op structure */
+       if (validate_dec_op(op) == -1) {
+               rte_bbdev_log(ERR, "Turbo decoder validation failed");
+               return -EINVAL;
+       }
+#endif
+
+       input = op->turbo_dec.input.data;
+       output_head = output = op->turbo_dec.hard_output.data;
+       mbuf_total_left = op->turbo_dec.input.length;
+       in_offset = op->turbo_dec.input.offset;
+       out_offset = op->turbo_dec.hard_output.offset;
+
+       if (!check_bit(op->turbo_dec.op_flags,
+               RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP))
+               crc24_overlap = 24;
+
+       c = op->turbo_dec.tb_params.c;
+       r = op->turbo_dec.tb_params.r;
+
+       while (mbuf_total_left > 0 && r < c && input != NULL) {
+               seg_total_left = rte_pktmbuf_data_len(input) - in_offset;
+               k = (r < op->turbo_dec.tb_params.c_neg) ?
+                               op->turbo_dec.tb_params.k_neg :
+                               op->turbo_dec.tb_params.k_pos;
+               kw = RTE_ALIGN_CEIL(k + 4, 32) * 3;
+
+               in_length = kw;
+               out_length = (k - crc24_overlap) >> 3;
+
+               mbuf_append(output_head, output, out_length);
+
+               if (seg_total_left < in_length) {
+                       rte_bbdev_log(ERR,
+                                       "Partial CB found in a TB. FPGA Driver doesn't support scatter-gather operations!");
+                       return -1;
+               }
+
+               /* Setup DMA Descriptor */
+               ring_offset = ((q->tail + desc_offset) & q->sw_ring_wrap_mask);
+               desc = q->ring_addr + ring_offset;
+               ret = fpga_dma_desc_td_fill(op, &desc->dec_req, input, output,
+                               in_length, k, in_offset, out_offset,
+                               ring_offset, cbs_in_op);
+               if (unlikely(ret < 0))
+                       return ret;
+
+               /* Update lengths */
+               ret = rte_pktmbuf_trim(op->turbo_dec.hard_output.data,
+                               (crc24_overlap >> 3));
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+               if (ret < 0) {
+                       rte_bbdev_log(ERR,
+                                       "The length to remove is greater than the length of the last segment");
+                       return -EINVAL;
+               }
+#endif
+               op->turbo_dec.hard_output.length += out_length;
+               mbuf_total_left -= in_length;
+
+               /* Update offsets */
+               if (seg_total_left == in_length) {
+                       /* Go to the next mbuf */
+                       input = input->next;
+                       output = output->next;
+                       in_offset = 0;
+                       out_offset = 0;
+               } else {
+                       in_offset += in_length;
+                       out_offset += out_length;
+               }
+
+               r++;
+               desc_offset++;
+               current_enqueued_cbs++;
+       }
+
+       if (mbuf_total_left > 0) {
+               rte_bbdev_log(ERR,
+                               "Some date still left for processing: mbuf_total_left = %u",
+                               mbuf_total_left);
+               return -1;
+       }
+
+       return current_enqueued_cbs;
+}
+
+static uint16_t
+fpga_enqueue_enc(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_enc_op **ops, uint16_t num)
+{
+       uint8_t cbs_in_op;
+       uint16_t i, total_enqueued_cbs = 0;
+       int32_t avail;
+       int enqueued_cbs;
+       struct fpga_queue *q = q_data->queue_private;
+       union fpga_dma_desc *desc;
+
+       /* Check if queue is not full */
+       if (unlikely(((q->tail + 1) & q->sw_ring_wrap_mask) ==
+                       q->head_free_desc))
+               return 0;
+
+       /* Calculates available space */
+       avail = (q->head_free_desc > q->tail) ?
+               q->head_free_desc - q->tail - 1 :
+               q->ring_ctrl_reg.ring_size + q->head_free_desc - q->tail - 1;
+
+       for (i = 0; i < num; ++i) {
+               if (ops[i]->turbo_enc.code_block_mode == 0) {
+                       cbs_in_op = get_num_cbs_in_op_enc(&ops[i]->turbo_enc);
+                       /* Check if there is available space for further
+                        * processing
+                        */
+                       if (unlikely(avail - cbs_in_op < 0))
+                               break;
+                       avail -= cbs_in_op;
+                       enqueued_cbs = enqueue_enc_one_op_tb(q, ops[i],
+                                       total_enqueued_cbs, cbs_in_op);
+               } else {
+                       /* Check if there is available space for further
+                        * processing
+                        */
+                       if (unlikely(avail - 1 < 0))
+                               break;
+                       avail -= 1;
+                       enqueued_cbs = enqueue_enc_one_op_cb(q, ops[i],
+                                       total_enqueued_cbs);
+               }
+
+               if (enqueued_cbs < 0)
+                       break;
+
+               total_enqueued_cbs += enqueued_cbs;
+
+               rte_bbdev_log_debug("enqueuing enc ops [%d/%d] | head %d | tail %d",
+                               total_enqueued_cbs, num,
+                               q->head_free_desc, q->tail);
+       }
+
+       /* Set interrupt bit for last CB in enqueued ops. FPGA issues interrupt
+        * only when all previous CBs were already processed.
+        */
+       desc = q->ring_addr + ((q->tail + total_enqueued_cbs - 1)
+                       & q->sw_ring_wrap_mask);
+       desc->enc_req.irq_en = q->irq_enable;
+
+       fpga_dma_enqueue(q, total_enqueued_cbs, &q_data->queue_stats);
+
+       /* Update stats */
+       q_data->queue_stats.enqueued_count += i;
+       q_data->queue_stats.enqueue_err_count += num - i;
+
+       return i;
+}
+
+static uint16_t
+fpga_enqueue_dec(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_dec_op **ops, uint16_t num)
+{
+       uint8_t cbs_in_op;
+       uint16_t i, total_enqueued_cbs = 0;
+       int32_t avail;
+       int enqueued_cbs;
+       struct fpga_queue *q = q_data->queue_private;
+       union fpga_dma_desc *desc;
+
+       /* Check if queue is not full */
+       if (unlikely(((q->tail + 1) & q->sw_ring_wrap_mask) ==
+                       q->head_free_desc))
+               return 0;
+
+       /* Calculates available space */
+       avail = (q->head_free_desc > q->tail) ?
+               q->head_free_desc - q->tail - 1 :
+               q->ring_ctrl_reg.ring_size + q->head_free_desc - q->tail - 1;
+
+       for (i = 0; i < num; ++i) {
+               if (ops[i]->turbo_dec.code_block_mode == 0) {
+                       cbs_in_op = get_num_cbs_in_op_dec(&ops[i]->turbo_dec);
+                       /* Check if there is available space for further
+                        * processing
+                        */
+                       if (unlikely(avail - cbs_in_op < 0))
+                               break;
+                       avail -= cbs_in_op;
+                       enqueued_cbs = enqueue_dec_one_op_tb(q, ops[i],
+                                       total_enqueued_cbs, cbs_in_op);
+               } else {
+                       /* Check if there is available space for further
+                        * processing
+                        */
+                       if (unlikely(avail - 1 < 0))
+                               break;
+                       avail -= 1;
+                       enqueued_cbs = enqueue_dec_one_op_cb(q, ops[i],
+                                       total_enqueued_cbs);
+               }
+
+               if (enqueued_cbs < 0)
+                       break;
+
+               total_enqueued_cbs += enqueued_cbs;
+
+               rte_bbdev_log_debug("enqueuing dec ops [%d/%d] | head %d | tail %d",
+                               total_enqueued_cbs, num,
+                               q->head_free_desc, q->tail);
+       }
+
+       /* Set interrupt bit for last CB in enqueued ops. FPGA issues interrupt
+        * only when all previous CBs were already processed.
+        */
+       desc = q->ring_addr + ((q->tail + total_enqueued_cbs - 1)
+                       & q->sw_ring_wrap_mask);
+       desc->dec_req.irq_en = q->irq_enable;
+
+       fpga_dma_enqueue(q, total_enqueued_cbs, &q_data->queue_stats);
+
+       /* Update stats */
+       q_data->queue_stats.enqueued_count += i;
+       q_data->queue_stats.enqueue_err_count += num - i;
+
+       return i;
+}
+
+static inline int
+dequeue_enc_one_op_cb(struct fpga_queue *q, struct rte_bbdev_enc_op **op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       int desc_error = 0;
+
+       /* Set current desc */
+       desc = q->ring_addr + ((q->head_free_desc + desc_offset)
+                       & q->sw_ring_wrap_mask);
+
+       /*check if done */
+       if (desc->enc_req.done == 0)
+               return -1;
+
+       /* make sure the response is read atomically */
+       rte_smp_rmb();
+
+       rte_bbdev_log_debug("DMA response desc %p", desc);
+
+       *op = desc->enc_req.op_addr;
+       /* Check the decriptor error field, return 1 on error */
+       desc_error = check_desc_error(desc->enc_req.error);
+       (*op)->status = desc_error << RTE_BBDEV_DATA_ERROR;
+
+       return 1;
+}
+
+static inline int
+dequeue_enc_one_op_tb(struct fpga_queue *q, struct rte_bbdev_enc_op **op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       uint8_t cbs_in_op, cb_idx;
+       int desc_error = 0;
+       int status = 0;
+
+       /* Set descriptor */
+       desc = q->ring_addr + ((q->head_free_desc + desc_offset)
+                       & q->sw_ring_wrap_mask);
+
+       /* Verify if done bit is set */
+       if (desc->enc_req.done == 0)
+               return -1;
+
+       /* Make sure the response is read atomically */
+       rte_smp_rmb();
+
+       /* Verify if done bit in all CBs is set */
+       cbs_in_op = desc->enc_req.cbs_in_op;
+       for (cb_idx = 1; cb_idx < cbs_in_op; ++cb_idx) {
+               desc = q->ring_addr + ((q->head_free_desc + desc_offset +
+                               cb_idx) & q->sw_ring_wrap_mask);
+               if (desc->enc_req.done == 0)
+                       return -1;
+       }
+
+       /* Make sure the response is read atomically */
+       rte_smp_rmb();
+
+       for (cb_idx = 0; cb_idx < cbs_in_op; ++cb_idx) {
+               desc = q->ring_addr + ((q->head_free_desc + desc_offset +
+                               cb_idx) & q->sw_ring_wrap_mask);
+               /* Check the decriptor error field, return 1 on error */
+               desc_error = check_desc_error(desc->enc_req.error);
+               status |=  desc_error << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log_debug("DMA response desc %p", desc);
+       }
+
+       *op = desc->enc_req.op_addr;
+       (*op)->status = status;
+       return cbs_in_op;
+}
+
+static inline int
+dequeue_dec_one_op_cb(struct fpga_queue *q, struct rte_bbdev_dec_op **op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       int desc_error = 0;
+       /* Set descriptor */
+       desc = q->ring_addr + ((q->head_free_desc + desc_offset)
+                       & q->sw_ring_wrap_mask);
+
+       /* Verify done bit is set */
+       if (desc->dec_req.done == 0)
+               return -1;
+
+       /* make sure the response is read atomically */
+       rte_smp_rmb();
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       print_dma_dec_desc_debug_info(desc);
+
+#endif
+
+       *op = desc->dec_req.op_addr;
+       /* FPGA reports in half-iterations, from 0 to 31. get ceiling */
+       (*op)->turbo_dec.iter_count = (desc->dec_req.iter + 2) >> 1;
+       /* crc_pass = 0 when decoder fails */
+       (*op)->status = !(desc->dec_req.crc_pass) << RTE_BBDEV_CRC_ERROR;
+       /* Check the decriptor error field, return 1 on error */
+       desc_error = check_desc_error(desc->enc_req.error);
+       (*op)->status |= desc_error << RTE_BBDEV_DATA_ERROR;
+       return 1;
+}
+
+static inline int
+dequeue_dec_one_op_tb(struct fpga_queue *q, struct rte_bbdev_dec_op **op,
+               uint16_t desc_offset)
+{
+       union fpga_dma_desc *desc;
+       uint8_t cbs_in_op, cb_idx, iter_count = 0;
+       int status = 0;
+       int  desc_error = 0;
+       /* Set descriptor */
+       desc = q->ring_addr + ((q->head_free_desc + desc_offset)
+                       & q->sw_ring_wrap_mask);
+
+       /* Verify if done bit is set */
+       if (desc->dec_req.done == 0)
+               return -1;
+
+       /* Make sure the response is read atomically */
+       rte_smp_rmb();
+
+       /* Verify if done bit in all CBs is set */
+       cbs_in_op = desc->dec_req.cbs_in_op;
+       for (cb_idx = 1; cb_idx < cbs_in_op; ++cb_idx) {
+               desc = q->ring_addr + ((q->head_free_desc + desc_offset +
+                               cb_idx) & q->sw_ring_wrap_mask);
+               if (desc->dec_req.done == 0)
+                       return -1;
+       }
+
+       /* Make sure the response is read atomically */
+       rte_smp_rmb();
+
+       for (cb_idx = 0; cb_idx < cbs_in_op; ++cb_idx) {
+               desc = q->ring_addr + ((q->head_free_desc + desc_offset +
+                               cb_idx) & q->sw_ring_wrap_mask);
+               /* get max iter_count for all CBs in op */
+               iter_count = RTE_MAX(iter_count, (uint8_t) desc->dec_req.iter);
+               /* crc_pass = 0 when decoder fails, one fails all */
+               status |= !(desc->dec_req.crc_pass) << RTE_BBDEV_CRC_ERROR;
+               /* Check the decriptor error field, return 1 on error */
+               desc_error = check_desc_error(desc->enc_req.error);
+               status |= desc_error << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log_debug("DMA response desc %p", desc);
+       }
+
+       *op = desc->dec_req.op_addr;
+
+       /* FPGA reports in half-iterations, get ceiling */
+       (*op)->turbo_dec.iter_count = (iter_count + 2) >> 1;
+       (*op)->status = status;
+       return cbs_in_op;
+}
+
+static uint16_t
+fpga_dequeue_enc(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_enc_op **ops, uint16_t num)
+{
+       struct fpga_queue *q = q_data->queue_private;
+       uint32_t avail = (q->tail - q->head_free_desc) & q->sw_ring_wrap_mask;
+       uint16_t i;
+       uint16_t dequeued_cbs = 0;
+       struct rte_bbdev_enc_op *op;
+       int ret;
+
+       for (i = 0; (i < num) && (dequeued_cbs < avail); ++i) {
+               op = (q->ring_addr + ((q->head_free_desc + dequeued_cbs)
+                       & q->sw_ring_wrap_mask))->enc_req.op_addr;
+               if (op->turbo_enc.code_block_mode == 0)
+                       ret = dequeue_enc_one_op_tb(q, &ops[i], dequeued_cbs);
+               else
+                       ret = dequeue_enc_one_op_cb(q, &ops[i], dequeued_cbs);
+
+               if (ret < 0)
+                       break;
+
+               dequeued_cbs += ret;
+
+               rte_bbdev_log_debug("dequeuing enc ops [%d/%d] | head %d | tail %d",
+                               dequeued_cbs, num, q->head_free_desc, q->tail);
+       }
+
+       /* Update head */
+       q->head_free_desc = (q->head_free_desc + dequeued_cbs) &
+                       q->sw_ring_wrap_mask;
+
+       /* Update stats */
+       q_data->queue_stats.dequeued_count += i;
+
+       return i;
+}
+
+static uint16_t
+fpga_dequeue_dec(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_dec_op **ops, uint16_t num)
+{
+       struct fpga_queue *q = q_data->queue_private;
+       uint32_t avail = (q->tail - q->head_free_desc) & q->sw_ring_wrap_mask;
+       uint16_t i;
+       uint16_t dequeued_cbs = 0;
+       struct rte_bbdev_dec_op *op;
+       int ret;
+
+       for (i = 0; (i < num) && (dequeued_cbs < avail); ++i) {
+               op = (q->ring_addr + ((q->head_free_desc + dequeued_cbs)
+                       & q->sw_ring_wrap_mask))->dec_req.op_addr;
+               if (op->turbo_dec.code_block_mode == 0)
+                       ret = dequeue_dec_one_op_tb(q, &ops[i], dequeued_cbs);
+               else
+                       ret = dequeue_dec_one_op_cb(q, &ops[i], dequeued_cbs);
+
+               if (ret < 0)
+                       break;
+
+               dequeued_cbs += ret;
+
+               rte_bbdev_log_debug("dequeuing dec ops [%d/%d] | head %d | tail %d",
+                               dequeued_cbs, num, q->head_free_desc, q->tail);
+       }
+
+       /* Update head */
+       q->head_free_desc = (q->head_free_desc + dequeued_cbs) &
+                       q->sw_ring_wrap_mask;
+
+       /* Update stats */
+       q_data->queue_stats.dequeued_count += i;
+
+       return i;
+}
+
+/* Initialization Function */
+static void
+fpga_lte_fec_init(struct rte_bbdev *dev)
+{
+       struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
+
+       dev->dev_ops = &fpga_ops;
+       dev->enqueue_enc_ops = fpga_enqueue_enc;
+       dev->enqueue_dec_ops = fpga_enqueue_dec;
+       dev->dequeue_enc_ops = fpga_dequeue_enc;
+       dev->dequeue_dec_ops = fpga_dequeue_dec;
+
+       ((struct fpga_lte_fec_device *) dev->data->dev_private)->pf_device =
+                       !strcmp(dev->device->driver->name,
+                                       RTE_STR(FPGA_LTE_FEC_PF_DRIVER_NAME));
+       ((struct fpga_lte_fec_device *) dev->data->dev_private)->mmio_base =
+                       pci_dev->mem_resource[0].addr;
+
+       rte_bbdev_log_debug(
+                       "Init device %s [%s] @ virtaddr %p phyaddr %#"PRIx64,
+                       dev->device->driver->name, dev->data->name,
+                       (void *)pci_dev->mem_resource[0].addr,
+                       pci_dev->mem_resource[0].phys_addr);
+}
+
+static int
+fpga_lte_fec_probe(struct rte_pci_driver *pci_drv __rte_unused,
+       struct rte_pci_device *pci_dev)
+{
+       struct rte_bbdev *bbdev = NULL;
+       char dev_name[RTE_BBDEV_NAME_MAX_LEN];
+
+       if (pci_dev == NULL) {
+               rte_bbdev_log(ERR, "NULL PCI device");
+               return -EINVAL;
+       }
+
+       rte_pci_device_name(&pci_dev->addr, dev_name, sizeof(dev_name));
+
+       /* Allocate memory to be used privately by drivers */
+       bbdev = rte_bbdev_allocate(pci_dev->device.name);
+       if (bbdev == NULL)
+               return -ENODEV;
+
+       /* allocate device private memory */
+       bbdev->data->dev_private = rte_zmalloc_socket(dev_name,
+                       sizeof(struct fpga_lte_fec_device), RTE_CACHE_LINE_SIZE,
+                       pci_dev->device.numa_node);
+
+       if (bbdev->data->dev_private == NULL) {
+               rte_bbdev_log(CRIT,
+                               "Allocate of %zu bytes for device \"%s\" failed",
+                               sizeof(struct fpga_lte_fec_device), dev_name);
+                               rte_bbdev_release(bbdev);
+                       return -ENOMEM;
+       }
+
+       /* Fill HW specific part of device structure */
+       bbdev->device = &pci_dev->device;
+       bbdev->intr_handle = &pci_dev->intr_handle;
+       bbdev->data->socket_id = pci_dev->device.numa_node;
+
+       /* Invoke FEC FPGA device initialization function */
+       fpga_lte_fec_init(bbdev);
+
+       rte_bbdev_log_debug("bbdev id = %u [%s]",
+                       bbdev->data->dev_id, dev_name);
+
+       struct fpga_lte_fec_device *d = bbdev->data->dev_private;
+       uint32_t version_id = fpga_reg_read_32(d->mmio_base,
+                       FPGA_LTE_FEC_VERSION_ID);
+       rte_bbdev_log(INFO, "FEC FPGA RTL v%u.%u",
+               ((uint16_t)(version_id >> 16)), ((uint16_t)version_id));
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       if (!strcmp(bbdev->device->driver->name,
+                       RTE_STR(FPGA_LTE_FEC_PF_DRIVER_NAME)))
+               print_static_reg_debug_info(d->mmio_base);
+#endif
+       return 0;
+}
+
+static int
+fpga_lte_fec_remove(struct rte_pci_device *pci_dev)
+{
+       struct rte_bbdev *bbdev;
+       int ret;
+       uint8_t dev_id;
+
+       if (pci_dev == NULL)
+               return -EINVAL;
+
+       /* Find device */
+       bbdev = rte_bbdev_get_named_dev(pci_dev->device.name);
+       if (bbdev == NULL) {
+               rte_bbdev_log(CRIT,
+                               "Couldn't find HW dev \"%s\" to uninitialise it",
+                               pci_dev->device.name);
+               return -ENODEV;
+       }
+       dev_id = bbdev->data->dev_id;
+
+       /* free device private memory before close */
+       rte_free(bbdev->data->dev_private);
+
+       /* Close device */
+       ret = rte_bbdev_close(dev_id);
+       if (ret < 0)
+               rte_bbdev_log(ERR,
+                               "Device %i failed to close during uninit: %i",
+                               dev_id, ret);
+
+       /* release bbdev from library */
+       ret = rte_bbdev_release(bbdev);
+       if (ret)
+               rte_bbdev_log(ERR, "Device %i failed to uninit: %i", dev_id,
+                               ret);
+
+       rte_bbdev_log_debug("Destroyed bbdev = %u", dev_id);
+
+       return 0;
+}
+
+static inline void
+set_default_fpga_conf(struct fpga_lte_fec_conf *def_conf)
+{
+       /* clear default configuration before initialization */
+       memset(def_conf, 0, sizeof(struct fpga_lte_fec_conf));
+       /* Set pf mode to true */
+       def_conf->pf_mode_en = true;
+
+       /* Set ratio between UL and DL to 1:1 (unit of weight is 3 CBs) */
+       def_conf->ul_bandwidth = 3;
+       def_conf->dl_bandwidth = 3;
+
+       /* Set Load Balance Factor to 64 */
+       def_conf->dl_load_balance = 64;
+       def_conf->ul_load_balance = 64;
+}
+
+/* Initial configuration of FPGA LTE FEC device */
+int
+fpga_lte_fec_configure(const char *dev_name,
+               const struct fpga_lte_fec_conf *conf)
+{
+       uint32_t payload_32, address;
+       uint16_t payload_16;
+       uint8_t payload_8;
+       uint16_t q_id, vf_id, total_q_id, total_ul_q_id, total_dl_q_id;
+       struct rte_bbdev *bbdev = rte_bbdev_get_named_dev(dev_name);
+       struct fpga_lte_fec_conf def_conf;
+
+       if (bbdev == NULL) {
+               rte_bbdev_log(ERR,
+                               "Invalid dev_name (%s), or device is not yet initialised",
+                               dev_name);
+               return -ENODEV;
+       }
+
+       struct fpga_lte_fec_device *d = bbdev->data->dev_private;
+
+       if (conf == NULL) {
+               rte_bbdev_log(ERR,
+                               "FPGA Configuration was not provided. Default configuration will be loaded.");
+               set_default_fpga_conf(&def_conf);
+               conf = &def_conf;
+       }
+
+       /*
+        * Configure UL:DL ratio.
+        * [7:0]: UL weight
+        * [15:8]: DL weight
+        */
+       payload_16 = (conf->dl_bandwidth << 8) | conf->ul_bandwidth;
+       address = FPGA_LTE_FEC_CONFIGURATION;
+       fpga_reg_write_16(d->mmio_base, address, payload_16);
+
+       /* Clear all queues registers */
+       payload_32 = FPGA_INVALID_HW_QUEUE_ID;
+       for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
+               address = (q_id << 2) + FPGA_LTE_FEC_QUEUE_MAP;
+               fpga_reg_write_32(d->mmio_base, address, payload_32);
+       }
+
+       /*
+        * If PF mode is enabled allocate all queues for PF only.
+        *
+        * For VF mode each VF can have different number of UL and DL queues.
+        * Total number of queues to configure cannot exceed FPGA
+        * capabilities - 64 queues - 32 queues for UL and 32 queues for DL.
+        * Queues mapping is done according to configuration:
+        *
+        * UL queues:
+        * |                Q_ID              | VF_ID |
+        * |                 0                |   0   |
+        * |                ...               |   0   |
+        * | conf->vf_dl_queues_number[0] - 1 |   0   |
+        * | conf->vf_dl_queues_number[0]     |   1   |
+        * |                ...               |   1   |
+        * | conf->vf_dl_queues_number[1] - 1 |   1   |
+        * |                ...               |  ...  |
+        * | conf->vf_dl_queues_number[7] - 1 |   7   |
+        *
+        * DL queues:
+        * |                Q_ID              | VF_ID |
+        * |                 32               |   0   |
+        * |                ...               |   0   |
+        * | conf->vf_ul_queues_number[0] - 1 |   0   |
+        * | conf->vf_ul_queues_number[0]     |   1   |
+        * |                ...               |   1   |
+        * | conf->vf_ul_queues_number[1] - 1 |   1   |
+        * |                ...               |  ...  |
+        * | conf->vf_ul_queues_number[7] - 1 |   7   |
+        *
+        * Example of configuration:
+        * conf->vf_ul_queues_number[0] = 4;  -> 4 UL queues for VF0
+        * conf->vf_dl_queues_number[0] = 4;  -> 4 DL queues for VF0
+        * conf->vf_ul_queues_number[1] = 2;  -> 2 UL queues for VF1
+        * conf->vf_dl_queues_number[1] = 2;  -> 2 DL queues for VF1
+        *
+        * UL:
+        * | Q_ID | VF_ID |
+        * |   0  |   0   |
+        * |   1  |   0   |
+        * |   2  |   0   |
+        * |   3  |   0   |
+        * |   4  |   1   |
+        * |   5  |   1   |
+        *
+        * DL:
+        * | Q_ID | VF_ID |
+        * |  32  |   0   |
+        * |  33  |   0   |
+        * |  34  |   0   |
+        * |  35  |   0   |
+        * |  36  |   1   |
+        * |  37  |   1   |
+        */
+       if (conf->pf_mode_en) {
+               payload_32 = 0x1;
+               for (q_id = 0; q_id < FPGA_TOTAL_NUM_QUEUES; ++q_id) {
+                       address = (q_id << 2) + FPGA_LTE_FEC_QUEUE_MAP;
+                       fpga_reg_write_32(d->mmio_base, address, payload_32);
+               }
+       } else {
+               /* Calculate total number of UL and DL queues to configure */
+               total_ul_q_id = total_dl_q_id = 0;
+               for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
+                       total_ul_q_id += conf->vf_ul_queues_number[vf_id];
+                       total_dl_q_id += conf->vf_dl_queues_number[vf_id];
+               }
+               total_q_id = total_dl_q_id + total_ul_q_id;
+               /*
+                * Check if total number of queues to configure does not exceed
+                * FPGA capabilities (64 queues - 32 UL and 32 DL queues)
+                */
+               if ((total_ul_q_id > FPGA_NUM_UL_QUEUES) ||
+                       (total_dl_q_id > FPGA_NUM_DL_QUEUES) ||
+                       (total_q_id > FPGA_TOTAL_NUM_QUEUES)) {
+                       rte_bbdev_log(ERR,
+                                       "FPGA Configuration failed. Too many queues to configure: UL_Q %u, DL_Q %u, FPGA_Q %u",
+                                       total_ul_q_id, total_dl_q_id,
+                                       FPGA_TOTAL_NUM_QUEUES);
+                       return -EINVAL;
+               }
+               total_ul_q_id = 0;
+               for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
+                       for (q_id = 0; q_id < conf->vf_ul_queues_number[vf_id];
+                                       ++q_id, ++total_ul_q_id) {
+                               address = (total_ul_q_id << 2) +
+                                               FPGA_LTE_FEC_QUEUE_MAP;
+                               payload_32 = ((0x80 + vf_id) << 16) | 0x1;
+                               fpga_reg_write_32(d->mmio_base, address,
+                                               payload_32);
+                       }
+               }
+               total_dl_q_id = 0;
+               for (vf_id = 0; vf_id < FPGA_LTE_FEC_NUM_VFS; ++vf_id) {
+                       for (q_id = 0; q_id < conf->vf_dl_queues_number[vf_id];
+                                       ++q_id, ++total_dl_q_id) {
+                               address = ((total_dl_q_id + FPGA_NUM_UL_QUEUES)
+                                               << 2) + FPGA_LTE_FEC_QUEUE_MAP;
+                               payload_32 = ((0x80 + vf_id) << 16) | 0x1;
+                               fpga_reg_write_32(d->mmio_base, address,
+                                               payload_32);
+                       }
+               }
+       }
+
+       /* Setting Load Balance Factor */
+       payload_16 = (conf->dl_load_balance << 8) | (conf->ul_load_balance);
+       address = FPGA_LTE_FEC_LOAD_BALANCE_FACTOR;
+       fpga_reg_write_16(d->mmio_base, address, payload_16);
+
+       /* Setting length of ring descriptor entry */
+       payload_16 = FPGA_RING_DESC_ENTRY_LENGTH;
+       address = FPGA_LTE_FEC_RING_DESC_LEN;
+       fpga_reg_write_16(d->mmio_base, address, payload_16);
+
+       /* Setting FLR timeout value */
+       payload_16 = conf->flr_time_out;
+       address = FPGA_LTE_FEC_FLR_TIME_OUT;
+       fpga_reg_write_16(d->mmio_base, address, payload_16);
+
+       /* Queue PF/VF mapping table is ready */
+       payload_8 = 0x1;
+       address = FPGA_LTE_FEC_QUEUE_PF_VF_MAP_DONE;
+       fpga_reg_write_8(d->mmio_base, address, payload_8);
+
+       rte_bbdev_log_debug("PF FPGA LTE FEC configuration complete for %s",
+                       dev_name);
+
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+       print_static_reg_debug_info(d->mmio_base);
+#endif
+       return 0;
+}
+
+/* FPGA LTE FEC PCI PF address map */
+static struct rte_pci_id pci_id_fpga_lte_fec_pf_map[] = {
+       {
+               RTE_PCI_DEVICE(FPGA_LTE_FEC_VENDOR_ID,
+                               FPGA_LTE_FEC_PF_DEVICE_ID)
+       },
+       {.device_id = 0},
+};
+
+static struct rte_pci_driver fpga_lte_fec_pci_pf_driver = {
+       .probe = fpga_lte_fec_probe,
+       .remove = fpga_lte_fec_remove,
+       .id_table = pci_id_fpga_lte_fec_pf_map,
+       .drv_flags = RTE_PCI_DRV_NEED_MAPPING
+};
+
+/* FPGA LTE FEC PCI VF address map */
+static struct rte_pci_id pci_id_fpga_lte_fec_vf_map[] = {
+       {
+               RTE_PCI_DEVICE(FPGA_LTE_FEC_VENDOR_ID,
+                               FPGA_LTE_FEC_VF_DEVICE_ID)
+       },
+       {.device_id = 0},
+};
+
+static struct rte_pci_driver fpga_lte_fec_pci_vf_driver = {
+       .probe = fpga_lte_fec_probe,
+       .remove = fpga_lte_fec_remove,
+       .id_table = pci_id_fpga_lte_fec_vf_map,
+       .drv_flags = RTE_PCI_DRV_NEED_MAPPING
+};
+
+
+RTE_PMD_REGISTER_PCI(FPGA_LTE_FEC_PF_DRIVER_NAME, fpga_lte_fec_pci_pf_driver);
+RTE_PMD_REGISTER_PCI_TABLE(FPGA_LTE_FEC_PF_DRIVER_NAME,
+               pci_id_fpga_lte_fec_pf_map);
+RTE_PMD_REGISTER_PCI(FPGA_LTE_FEC_VF_DRIVER_NAME, fpga_lte_fec_pci_vf_driver);
+RTE_PMD_REGISTER_PCI_TABLE(FPGA_LTE_FEC_VF_DRIVER_NAME,
+               pci_id_fpga_lte_fec_vf_map);
+
+RTE_INIT(fpga_lte_fec_init_log)
+{
+       fpga_lte_fec_logtype = rte_log_register("pmd.bb.fpga_lte_fec");
+       if (fpga_lte_fec_logtype >= 0)
+#ifdef RTE_LIBRTE_BBDEV_DEBUG
+               rte_log_set_level(fpga_lte_fec_logtype, RTE_LOG_DEBUG);
+#else
+               rte_log_set_level(fpga_lte_fec_logtype, RTE_LOG_NOTICE);
+#endif
+}
diff --git a/drivers/baseband/fpga_lte_fec/fpga_lte_fec.h b/drivers/baseband/fpga_lte_fec/fpga_lte_fec.h
new file mode 100644 (file)
index 0000000..9ae8b12
--- /dev/null
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Intel Corporation
+ */
+
+#ifndef _FPGA_LTE_FEC_H_
+#define _FPGA_LTE_FEC_H_
+
+#include <stdint.h>
+#include <stdbool.h>
+
+/**
+ * @file fpga_lte_fec.h
+ *
+ * Interface for Intel(R) FGPA LTE FEC device configuration at the host level,
+ * directly accessible by the application.
+ * Configuration related to LTE Turbo coding functionality is done through
+ * librte_bbdev library.
+ *
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice
+ */
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**< Number of Virtual Functions FGPA 4G FEC supports */
+#define FPGA_LTE_FEC_NUM_VFS 8
+
+/**
+ * Structure to pass FPGA 4G FEC configuration.
+ */
+struct fpga_lte_fec_conf {
+       /**< 1 if PF is used for dataplane, 0 for VFs */
+       bool pf_mode_en;
+       /**< Number of UL queues per VF */
+       uint8_t vf_ul_queues_number[FPGA_LTE_FEC_NUM_VFS];
+       /**< Number of DL queues per VF */
+       uint8_t vf_dl_queues_number[FPGA_LTE_FEC_NUM_VFS];
+       /**< UL bandwidth. Needed for schedule algorithm */
+       uint8_t ul_bandwidth;
+       /**< DL bandwidth. Needed for schedule algorithm */
+       uint8_t dl_bandwidth;
+       /**< UL Load Balance */
+       uint8_t ul_load_balance;
+       /**< DL Load Balance */
+       uint8_t dl_load_balance;
+       /**< FLR timeout value */
+       uint16_t flr_time_out;
+};
+
+/**
+ * Configure Intel(R) FPGA LTE FEC device
+ *
+ * @param dev_name
+ *   The name of the device. This is the short form of PCI BDF, e.g. 00:01.0.
+ *   It can also be retrieved for a bbdev device from the dev_name field in the
+ *   rte_bbdev_info structure returned by rte_bbdev_info_get().
+ * @param conf
+ *   Configuration to apply to FPGA 4G FEC.
+ *
+ * @return
+ *   Zero on success, negative value on failure.
+ */
+int
+fpga_lte_fec_configure(const char *dev_name,
+               const struct fpga_lte_fec_conf *conf);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _FPGA_LTE_FEC_H_ */
diff --git a/drivers/baseband/fpga_lte_fec/meson.build b/drivers/baseband/fpga_lte_fec/meson.build
new file mode 100644 (file)
index 0000000..bf44e6b
--- /dev/null
@@ -0,0 +1,7 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2019 Intel Corporation
+
+deps += ['bbdev', 'bus_vdev', 'ring', 'pci', 'bus_pci']
+name = 'bbdev_fpga_lte_fec'
+allow_experimental_apis = true
+sources = files('fpga_lte_fec.c')
diff --git a/drivers/baseband/fpga_lte_fec/rte_pmd_bbdev_fpga_lte_fec_version.map b/drivers/baseband/fpga_lte_fec/rte_pmd_bbdev_fpga_lte_fec_version.map
new file mode 100644 (file)
index 0000000..e923270
--- /dev/null
@@ -0,0 +1,3 @@
+DPDK_19.08 {
+    local: *;
+};
index 40a87d2..391bd51 100644 (file)
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2018 Luca Boccassi <bluca@debian.org>
 
-drivers = ['null', 'turbo_sw']
+drivers = ['null', 'turbo_sw', 'fpga_lte_fec']
 
 config_flag_fmt = 'RTE_LIBRTE_@0@_PMD'
 driver_name_fmt = 'rte_pmd_@0@'
index 5dcef13..8bda05a 100644 (file)
@@ -226,6 +226,7 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD)     += -lrte_pmd_netvsc
 
 ifeq ($(CONFIG_RTE_LIBRTE_BBDEV),y)
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_NULL)     += -lrte_pmd_bbdev_null
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_FPGA_LTE_FEC) += -lrte_pmd_fpga_lte_fec
 
 # TURBO SOFTWARE PMD is dependent on the FLEXRAN library
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -lrte_pmd_bbdev_turbo_sw