bb/turbo_sw: add software turbo driver
authorAmr Mokhtar <amr.mokhtar@intel.com>
Thu, 11 Jan 2018 19:23:20 +0000 (19:23 +0000)
committerThomas Monjalon <thomas@monjalon.net>
Fri, 19 Jan 2018 00:44:25 +0000 (01:44 +0100)
- bbdev 'turbo_sw' is the software accelerated version of 3GPP L1
 Turbo coding operation using the optimized Intel FlexRAN SDK libraries.
- 'turbo_sw' pmd is disabled by default

Signed-off-by: Amr Mokhtar <amr.mokhtar@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
config/common_base
devtools/test-build.sh
doc/guides/bbdevs/index.rst
doc/guides/bbdevs/turbo_sw.rst [new file with mode: 0644]
drivers/bbdev/Makefile
drivers/bbdev/turbo_sw/Makefile [new file with mode: 0644]
drivers/bbdev/turbo_sw/bbdev_turbo_software.c [new file with mode: 0644]
drivers/bbdev/turbo_sw/rte_pmd_bbdev_turbo_sw_version.map [new file with mode: 0644]
mk/rte.app.mk

index bca1880..fe0bbc2 100644 (file)
@@ -457,6 +457,11 @@ CONFIG_RTE_BBDEV_MAX_DEVS=128
 #
 CONFIG_RTE_LIBRTE_PMD_BBDEV_NULL=y
 
+#
+# Compile PMD for turbo software bbdev device
+#
+CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW=n
+
 #
 # Compile generic crypto device library
 #
index 77a2a47..3362edc 100755 (executable)
@@ -47,6 +47,7 @@ default_path=$PATH
 # - DPDK_DEP_ZLIB (y/[n])
 # - DPDK_MAKE_JOBS (int)
 # - DPDK_NOTIFY (notify-send)
+# - FLEXRAN_SDK
 # - LIBMUSDK_PATH
 # - LIBSSO_SNOW3G_PATH
 # - LIBSSO_KASUMI_PATH
@@ -130,6 +131,7 @@ reset_env ()
        unset DPDK_DEP_ZLIB
        unset AESNI_MULTI_BUFFER_LIB_PATH
        unset ARMV8_CRYPTO_LIB_PATH
+       unset FLEXRAN_SDK
        unset LIBMUSDK_PATH
        unset LIBSSO_SNOW3G_PATH
        unset LIBSSO_KASUMI_PATH
@@ -194,6 +196,8 @@ config () # <directory> <target> <options>
                sed -ri        's,(PMD_OPENSSL=)n,\1y,' $1/.config
                test "$DPDK_DEP_SSL" != y || \
                sed -ri            's,(PMD_QAT=)n,\1y,' $1/.config
+               test -z "$FLEXRAN_SDK" || \
+               sed -ri     's,(BBDEV_TURBO_SW=)n,\1y,' $1/.config
                sed -ri           's,(SCHED_.*=)n,\1y,' $1/.config
                test -z "$LIBMUSDK_PATH" || \
                sed -ri    's,(PMD_MRVL_CRYPTO=)n,\1y,' $1/.config
index 5cd411b..93276ed 100644 (file)
@@ -9,3 +9,4 @@ Baseband Device Drivers
     :numbered:
 
     null
+    turbo_sw
diff --git a/doc/guides/bbdevs/turbo_sw.rst b/doc/guides/bbdevs/turbo_sw.rst
new file mode 100644 (file)
index 0000000..b3fed16
--- /dev/null
@@ -0,0 +1,147 @@
+..  SPDX-License-Identifier: BSD-3-Clause
+    Copyright(c) 2017 Intel Corporation
+
+SW Turbo Poll Mode Driver
+=========================
+
+The SW Turbo PMD (**turbo_sw**) provides a poll mode bbdev driver that utilizes
+Intel optimized libraries for LTE Layer 1 workloads acceleration. This PMD
+supports the functions: Turbo FEC, Rate Matching and CRC functions.
+
+Features
+--------
+
+SW Turbo PMD has support for the following capabilities:
+
+For the encode operation:
+
+* ``RTE_BBDEV_TURBO_CRC_24A_ATTACH``
+* ``RTE_BBDEV_TURBO_CRC_24B_ATTACH``
+* ``RTE_BBDEV_TURBO_RATE_MATCH``
+* ``RTE_BBDEV_TURBO_RV_INDEX_BYPASS``
+
+For the decode operation:
+
+* ``RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE``
+* ``RTE_BBDEV_TURBO_CRC_TYPE_24B``
+* ``RTE_BBDEV_TURBO_POS_LLR_1_BIT_IN``
+* ``RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN``
+
+
+Limitations
+-----------
+
+* In-place operations for Turbo encode and decode are not supported
+
+Installation
+------------
+
+FlexRAN SDK Download
+~~~~~~~~~~~~~~~~~~~~
+
+To build DPDK with the *turbo_sw* PMD the user is required to download
+the export controlled ``FlexRAN SDK`` Libraries. An account at Intel Resource
+Design Center needs to be registered from
+`<https://www.intel.com/content/www/us/en/design/resource-design-center.html>`_.
+
+Once registered, the user needs to log in, and look for
+*Intel SWA_SW_FlexRAN_Release_Package R1_3_0* and click for download. Or use
+this direct download link `<https://cdrd.intel.com/v1/dl/getContent/575367>`_.
+
+After download is complete, the user needs to unpack and compile on their
+system before building DPDK.
+
+FlexRAN SDK Installation
+~~~~~~~~~~~~~~~~~~~~~~~~
+
+The following are pre-requisites for building FlexRAN SDK Libraries:
+ (a) An AVX2 supporting machine
+ (b) Windriver TS 2 or CentOS 7 operating systems
+ (c) Intel ICC compiler installed
+
+The following instructions should be followed in this exact order:
+
+#. Set the environment variables:
+
+    .. code-block:: console
+
+        source <path-to-icc-compiler-install-folder>/linux/bin/compilervars.sh intel64 -platform linux
+
+
+#. Extract the ``FlexRAN-1.3.0.tar.gz.zip`` package, then run the SDK extractor
+   script and accept the license:
+
+    .. code-block:: console
+
+        cd <path-to-workspace>/FlexRAN-1.3.0/
+        ./SDK-R1.3.0.sh
+
+#. To allow ``FlexRAN SDK R1.3.0`` to work with bbdev properly, the following
+   hotfix is required. Change the return of function ``rate_matching_turbo_lte_avx2()``
+   located in file
+   ``<path-to-workspace>/FlexRAN-1.3.0/SDK-R1.3.0/sdk/source/phy/lib_rate_matching/phy_rate_match_avx2.cpp``
+   to return 0 instead of 1.
+
+    .. code-block:: c
+
+        -  return 1;
+        +  return 0;
+
+#. Generate makefiles based on system configuration:
+
+    .. code-block:: console
+
+        cd <path-to-workspace>/FlexRAN-1.3.0/SDK-R1.3.0/sdk/
+        ./create-makefiles-linux.sh
+
+#. A build folder is generated in this form ``build-<ISA>-<CC>``, enter that
+   folder and install:
+
+    .. code-block:: console
+
+        cd build-avx2-icc/
+        make install
+
+
+Initialization
+--------------
+
+In order to enable this virtual bbdev PMD, the user must:
+
+* Build the ``FLEXRAN SDK`` libraries (explained in Installation section).
+
+* Export the environmental variables ``FLEXRAN_SDK`` to the path where the
+  FlexRAN SDK libraries were installed. And ``DIR_WIRELESS_SDK`` to the path
+  where the libraries were extracted.
+
+Example:
+
+.. code-block:: console
+
+    export FLEXRAN_SDK=<path-to-workspace>/FlexRAN-1.3.0/SDK-R1.3.0/sdk/build-avx2-icc/install
+    export DIR_WIRELESS_SDK=<path-to-workspace>/FlexRAN-1.3.0/SDK-R1.3.0/sdk/
+
+
+* Set ``CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW=y`` in DPDK common configuration
+  file ``config/common_base``.
+
+To use the PMD in an application, user must:
+
+- Call ``rte_vdev_init("turbo_sw")`` within the application.
+
+- Use ``--vdev="turbo_sw"`` in the EAL options, which will call ``rte_vdev_init()`` internally.
+
+The following parameters (all optional) can be provided in the previous two calls:
+
+* ``socket_id``: Specify the socket where the memory for the device is going to be allocated
+  (by default, *socket_id* will be the socket where the core that is creating the PMD is running on).
+
+* ``max_nb_queues``: Specify the maximum number of queues in the device (default is ``RTE_MAX_LCORE``).
+
+Example:
+~~~~~~~~
+
+.. code-block:: console
+
+    ./test-bbdev.py -e="--vdev=turbo_sw,socket_id=0,max_nb_queues=8" \
+    -c validation -v ./test_vectors/bbdev_vector_t?_default.data
index 7611874..4ec83b0 100644 (file)
@@ -8,5 +8,7 @@ core-libs += librte_bbdev librte_kvargs librte_cfgfile
 
 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)
 
 include $(RTE_SDK)/mk/rte.subdir.mk
diff --git a/drivers/bbdev/turbo_sw/Makefile b/drivers/bbdev/turbo_sw/Makefile
new file mode 100644 (file)
index 0000000..08f24b0
--- /dev/null
@@ -0,0 +1,41 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright(c) 2017 Intel Corporation
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+ifeq ($(FLEXRAN_SDK),)
+$(error "Please define FLEXRAN_SDK environment variable")
+endif
+
+# library name
+LIB = librte_pmd_bbdev_turbo_sw.a
+
+# build flags
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring -lrte_kvargs
+LDLIBS += -lrte_bbdev
+LDLIBS += -lrte_bus_vdev
+
+# versioning export map
+EXPORT_MAP := rte_pmd_bbdev_turbo_sw_version.map
+
+# external library dependencies
+CFLAGS += -I$(FLEXRAN_SDK)/lib_common
+CFLAGS += -I$(FLEXRAN_SDK)/lib_turbo
+CFLAGS += -I$(FLEXRAN_SDK)/lib_crc
+CFLAGS += -I$(FLEXRAN_SDK)/lib_rate_matching
+
+LDLIBS += -L$(FLEXRAN_SDK)/lib_crc -lcrc
+LDLIBS += -L$(FLEXRAN_SDK)/lib_turbo -lturbo
+LDLIBS += -L$(FLEXRAN_SDK)/lib_rate_matching -lrate_matching
+LDLIBS += -L$(FLEXRAN_SDK)/lib_common -lcommon
+LDLIBS += -lstdc++ -lirc -limf -lipps
+
+# library version
+LIBABIVER := 1
+
+# library source files
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += bbdev_turbo_software.c
+
+include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/drivers/bbdev/turbo_sw/bbdev_turbo_software.c b/drivers/bbdev/turbo_sw/bbdev_turbo_software.c
new file mode 100644 (file)
index 0000000..981da6e
--- /dev/null
@@ -0,0 +1,1206 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Intel Corporation
+ */
+
+#include <string.h>
+
+#include <rte_common.h>
+#include <rte_bus_vdev.h>
+#include <rte_malloc.h>
+#include <rte_ring.h>
+#include <rte_kvargs.h>
+
+#include <rte_bbdev.h>
+#include <rte_bbdev_pmd.h>
+
+#include <phy_turbo.h>
+#include <phy_crc.h>
+#include <phy_rate_match.h>
+#include <divide.h>
+
+#define DRIVER_NAME turbo_sw
+
+/* Number of columns in sub-block interleaver (36.212, section 5.1.4.1.1) */
+#define C_SUBBLOCK (32)
+#define MAX_TB_SIZE (391656)
+#define MAX_CB_SIZE (6144)
+#define MAX_KW (18528)
+
+/* private data structure */
+struct bbdev_private {
+       unsigned int max_nb_queues;  /**< Max number of queues */
+};
+
+/*  Initialisation params structure that can be used by Turbo SW driver */
+struct turbo_sw_params {
+       int socket_id;  /*< Turbo SW device socket */
+       uint16_t queues_num;  /*< Turbo SW device queues number */
+};
+
+/* Accecptable params for Turbo SW devices */
+#define TURBO_SW_MAX_NB_QUEUES_ARG  "max_nb_queues"
+#define TURBO_SW_SOCKET_ID_ARG      "socket_id"
+
+static const char * const turbo_sw_valid_params[] = {
+       TURBO_SW_MAX_NB_QUEUES_ARG,
+       TURBO_SW_SOCKET_ID_ARG
+};
+
+/* queue */
+struct turbo_sw_queue {
+       /* Ring for processed (encoded/decoded) operations which are ready to
+        * be dequeued.
+        */
+       struct rte_ring *processed_pkts;
+       /* Stores input for turbo encoder (used when CRC attachment is
+        * performed
+        */
+       uint8_t *enc_in;
+       /* Stores output from turbo encoder */
+       uint8_t *enc_out;
+       /* Alpha gamma buf for bblib_turbo_decoder() function */
+       int8_t *ag;
+       /* Temp buf for bblib_turbo_decoder() function */
+       uint16_t *code_block;
+       /* Input buf for bblib_rate_dematching_lte() function */
+       uint8_t *deint_input;
+       /* Output buf for bblib_rate_dematching_lte() function */
+       uint8_t *deint_output;
+       /* Output buf for bblib_turbodec_adapter_lte() function */
+       uint8_t *adapter_output;
+       /* Operation type of this queue */
+       enum rte_bbdev_op_type type;
+} __rte_cache_aligned;
+
+/* Calculate index based on Table 5.1.3-3 from TS34.212 */
+static inline int32_t
+compute_idx(uint16_t k)
+{
+       int32_t result = 0;
+
+       if (k < 40 || k > MAX_CB_SIZE)
+               return -1;
+
+       if (k > 2048) {
+               if ((k - 2048) % 64 != 0)
+                       result = -1;
+
+               result = 124 + (k - 2048) / 64;
+       } else if (k <= 512) {
+               if ((k - 40) % 8 != 0)
+                       result = -1;
+
+               result = (k - 40) / 8 + 1;
+       } else if (k <= 1024) {
+               if ((k - 512) % 16 != 0)
+                       result = -1;
+
+               result = 60 + (k - 512) / 16;
+       } else { /* 1024 < k <= 2048 */
+               if ((k - 1024) % 32 != 0)
+                       result = -1;
+
+               result = 92 + (k - 1024) / 32;
+       }
+
+       return result;
+}
+
+/* Read flag value 0/1 from bitmap */
+static inline bool
+check_bit(uint32_t bitmap, uint32_t bitmask)
+{
+       return bitmap & bitmask;
+}
+
+/* Get device info */
+static void
+info_get(struct rte_bbdev *dev, struct rte_bbdev_driver_info *dev_info)
+{
+       struct bbdev_private *internals = dev->data->dev_private;
+
+       static const struct rte_bbdev_op_cap bbdev_capabilities[] = {
+               {
+                       .type = RTE_BBDEV_OP_TURBO_DEC,
+                       .cap.turbo_dec = {
+                               .capability_flags =
+                                       RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE |
+                                       RTE_BBDEV_TURBO_POS_LLR_1_BIT_IN |
+                                       RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN |
+                                       RTE_BBDEV_TURBO_CRC_TYPE_24B |
+                                       RTE_BBDEV_TURBO_EARLY_TERMINATION,
+                               .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_CRC_24A_ATTACH |
+                                               RTE_BBDEV_TURBO_RATE_MATCH |
+                                               RTE_BBDEV_TURBO_RV_INDEX_BYPASS,
+                               .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 = {
+               .queue_size = RTE_BBDEV_QUEUE_SIZE_LIMIT,
+       };
+
+       static const enum rte_cpu_flag_t cpu_flag = RTE_CPUFLAG_SSE4_2;
+
+       default_queue_conf.socket = dev->data->socket_id;
+
+       dev_info->driver_name = RTE_STR(DRIVER_NAME);
+       dev_info->max_num_queues = internals->max_nb_queues;
+       dev_info->queue_size_lim = RTE_BBDEV_QUEUE_SIZE_LIMIT;
+       dev_info->hardware_accelerated = false;
+       dev_info->max_queue_priority = 0;
+       dev_info->default_queue_conf = default_queue_conf;
+       dev_info->capabilities = bbdev_capabilities;
+       dev_info->cpu_flag_reqs = &cpu_flag;
+       dev_info->min_alignment = 64;
+
+       rte_bbdev_log_debug("got device info from %u\n", dev->data->dev_id);
+}
+
+/* Release queue */
+static int
+q_release(struct rte_bbdev *dev, uint16_t q_id)
+{
+       struct turbo_sw_queue *q = dev->data->queues[q_id].queue_private;
+
+       if (q != NULL) {
+               rte_ring_free(q->processed_pkts);
+               rte_free(q->enc_out);
+               rte_free(q->enc_in);
+               rte_free(q->ag);
+               rte_free(q->code_block);
+               rte_free(q->deint_input);
+               rte_free(q->deint_output);
+               rte_free(q->adapter_output);
+               rte_free(q);
+               dev->data->queues[q_id].queue_private = NULL;
+       }
+
+       rte_bbdev_log_debug("released device queue %u:%u",
+                       dev->data->dev_id, q_id);
+       return 0;
+}
+
+/* Setup a queue */
+static int
+q_setup(struct rte_bbdev *dev, uint16_t q_id,
+               const struct rte_bbdev_queue_conf *queue_conf)
+{
+       int ret;
+       struct turbo_sw_queue *q;
+       char name[RTE_RING_NAMESIZE];
+
+       /* Allocate the queue data structure. */
+       q = rte_zmalloc_socket(RTE_STR(DRIVER_NAME), sizeof(*q),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q == NULL) {
+               rte_bbdev_log(ERR, "Failed to allocate queue memory");
+               return -ENOMEM;
+       }
+
+       /* Allocate memory for encoder output. */
+       ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_enc_out%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->enc_out = rte_zmalloc_socket(name,
+                       ((MAX_TB_SIZE >> 3) + 3) * sizeof(*q->enc_out) * 3,
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->enc_out == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for rate matching output. */
+       ret = snprintf(name, RTE_RING_NAMESIZE,
+                       RTE_STR(DRIVER_NAME)"_enc_in%u:%u", dev->data->dev_id,
+                       q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->enc_in = rte_zmalloc_socket(name,
+                       (MAX_CB_SIZE >> 3) * sizeof(*q->enc_in),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->enc_in == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for Aplha Gamma temp buffer. */
+       ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_ag%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->ag = rte_zmalloc_socket(name,
+                       MAX_CB_SIZE * 10 * sizeof(*q->ag),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->ag == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for code block temp buffer. */
+       ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"_cb%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->code_block = rte_zmalloc_socket(name,
+                       (6144 >> 3) * sizeof(*q->code_block),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->code_block == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for Deinterleaver input. */
+       ret = snprintf(name, RTE_RING_NAMESIZE,
+                       RTE_STR(DRIVER_NAME)"_deint_input%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->deint_input = rte_zmalloc_socket(name,
+                       MAX_KW * sizeof(*q->deint_input),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->deint_input == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for Deinterleaver output. */
+       ret = snprintf(name, RTE_RING_NAMESIZE,
+                       RTE_STR(DRIVER_NAME)"_deint_output%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->deint_output = rte_zmalloc_socket(NULL,
+                       MAX_KW * sizeof(*q->deint_output),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->deint_output == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Allocate memory for Adapter output. */
+       ret = snprintf(name, RTE_RING_NAMESIZE,
+                       RTE_STR(DRIVER_NAME)"_adapter_output%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->adapter_output = rte_zmalloc_socket(NULL,
+                       MAX_CB_SIZE * 6 * sizeof(*q->adapter_output),
+                       RTE_CACHE_LINE_SIZE, queue_conf->socket);
+       if (q->adapter_output == NULL) {
+               rte_bbdev_log(ERR,
+                       "Failed to allocate queue memory for %s", name);
+               goto free_q;
+       }
+
+       /* Create ring for packets awaiting to be dequeued. */
+       ret = snprintf(name, RTE_RING_NAMESIZE, RTE_STR(DRIVER_NAME)"%u:%u",
+                       dev->data->dev_id, q_id);
+       if ((ret < 0) || (ret >= (int)RTE_RING_NAMESIZE)) {
+               rte_bbdev_log(ERR,
+                               "Creating queue name for device %u queue %u failed",
+                               dev->data->dev_id, q_id);
+               return -ENAMETOOLONG;
+       }
+       q->processed_pkts = rte_ring_create(name, queue_conf->queue_size,
+                       queue_conf->socket, RING_F_SP_ENQ | RING_F_SC_DEQ);
+       if (q->processed_pkts == NULL) {
+               rte_bbdev_log(ERR, "Failed to create ring for %s", name);
+               goto free_q;
+       }
+
+       q->type = queue_conf->op_type;
+
+       dev->data->queues[q_id].queue_private = q;
+       rte_bbdev_log_debug("setup device queue %s", name);
+       return 0;
+
+free_q:
+       rte_ring_free(q->processed_pkts);
+       rte_free(q->enc_out);
+       rte_free(q->enc_in);
+       rte_free(q->ag);
+       rte_free(q->code_block);
+       rte_free(q->deint_input);
+       rte_free(q->deint_output);
+       rte_free(q->adapter_output);
+       rte_free(q);
+       return -EFAULT;
+}
+
+static const struct rte_bbdev_ops pmd_ops = {
+       .info_get = info_get,
+       .queue_setup = q_setup,
+       .queue_release = q_release
+};
+
+/* Checks if the encoder input buffer is correct.
+ * Returns 0 if it's valid, -1 otherwise.
+ */
+static inline int
+is_enc_input_valid(const uint16_t k, const int32_t k_idx,
+               const uint16_t in_length)
+{
+       if (k_idx < 0) {
+               rte_bbdev_log(ERR, "K Index is invalid");
+               return -1;
+       }
+
+       if (in_length - (k >> 3) < 0) {
+               rte_bbdev_log(ERR,
+                               "Mismatch between input length (%u bytes) and K (%u bits)",
+                               in_length, k);
+               return -1;
+       }
+
+       if (k > MAX_CB_SIZE) {
+               rte_bbdev_log(ERR, "CB size (%u) is too big, max: %d",
+                               k, MAX_CB_SIZE);
+               return -1;
+       }
+
+       return 0;
+}
+
+/* Checks if the decoder input buffer is correct.
+ * Returns 0 if it's valid, -1 otherwise.
+ */
+static inline int
+is_dec_input_valid(int32_t k_idx, int16_t kw, int16_t in_length)
+{
+       if (k_idx < 0) {
+               rte_bbdev_log(ERR, "K index is invalid");
+               return -1;
+       }
+
+       if (in_length - kw < 0) {
+               rte_bbdev_log(ERR,
+                               "Mismatch between input length (%u) and kw (%u)",
+                               in_length, kw);
+               return -1;
+       }
+
+       if (kw > MAX_KW) {
+               rte_bbdev_log(ERR, "Input length (%u) is too big, max: %d",
+                               kw, MAX_KW);
+               return -1;
+       }
+
+       return 0;
+}
+
+static inline void
+process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
+               uint8_t cb_idx, uint8_t c, uint16_t k, uint16_t ncb,
+               uint32_t e, struct rte_mbuf *m_in, struct rte_mbuf *m_out,
+               uint16_t in_offset, uint16_t out_offset, uint16_t total_left)
+{
+       int ret;
+       int16_t k_idx;
+       uint16_t m;
+       uint8_t *in, *out0, *out1, *out2, *tmp_out, *rm_out;
+       struct rte_bbdev_op_turbo_enc *enc = &op->turbo_enc;
+       struct bblib_crc_request crc_req;
+       struct bblib_turbo_encoder_request turbo_req;
+       struct bblib_turbo_encoder_response turbo_resp;
+       struct bblib_rate_match_dl_request rm_req;
+       struct bblib_rate_match_dl_response rm_resp;
+
+       k_idx = compute_idx(k);
+       in = rte_pktmbuf_mtod_offset(m_in, uint8_t *, in_offset);
+
+       /* CRC24A (for TB) */
+       if ((enc->op_flags & RTE_BBDEV_TURBO_CRC_24A_ATTACH) &&
+               (enc->code_block_mode == 1)) {
+               ret = is_enc_input_valid(k - 24, k_idx, total_left);
+               if (ret != 0) {
+                       op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+                       return;
+               }
+               /* copy the input to the temporary buffer to be able to extend
+                * it by 3 CRC bytes
+                */
+               rte_memcpy(q->enc_in, in, (k - 24) >> 3);
+               crc_req.data = q->enc_in;
+               crc_req.len = (k - 24) >> 3;
+               if (bblib_lte_crc24a_gen(&crc_req) == -1) {
+                       op->status |= 1 << RTE_BBDEV_CRC_ERROR;
+                       rte_bbdev_log(ERR, "CRC24a generation failed");
+                       return;
+               }
+               in = q->enc_in;
+       } else if (enc->op_flags & RTE_BBDEV_TURBO_CRC_24B_ATTACH) {
+               /* CRC24B */
+               ret = is_enc_input_valid(k - 24, k_idx, total_left);
+               if (ret != 0) {
+                       op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+                       return;
+               }
+               /* copy the input to the temporary buffer to be able to extend
+                * it by 3 CRC bytes
+                */
+               rte_memcpy(q->enc_in, in, (k - 24) >> 3);
+               crc_req.data = q->enc_in;
+               crc_req.len = (k - 24) >> 3;
+               if (bblib_lte_crc24b_gen(&crc_req) == -1) {
+                       op->status |= 1 << RTE_BBDEV_CRC_ERROR;
+                       rte_bbdev_log(ERR, "CRC24b generation failed");
+                       return;
+               }
+               in = q->enc_in;
+       } else {
+               ret = is_enc_input_valid(k, k_idx, total_left);
+               if (ret != 0) {
+                       op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+                       return;
+               }
+       }
+
+       /* Turbo encoder */
+
+       /* Each bit layer output from turbo encoder is (k+4) bits long, i.e.
+        * input length + 4 tail bits. That's (k/8) + 1 bytes after rounding up.
+        * So dst_data's length should be 3*(k/8) + 3 bytes.
+        */
+       out0 = q->enc_out;
+       out1 = RTE_PTR_ADD(out0, (k >> 3) + 1);
+       out2 = RTE_PTR_ADD(out1, (k >> 3) + 1);
+
+       turbo_req.case_id = k_idx;
+       turbo_req.input_win = in;
+       turbo_req.length = k >> 3;
+       turbo_resp.output_win_0 = out0;
+       turbo_resp.output_win_1 = out1;
+       turbo_resp.output_win_2 = out2;
+       if (bblib_turbo_encoder(&turbo_req, &turbo_resp) != 0) {
+               op->status |= 1 << RTE_BBDEV_DRV_ERROR;
+               rte_bbdev_log(ERR, "Turbo Encoder failed");
+               return;
+       }
+
+       /* Rate-matching */
+       if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH) {
+               /* get output data starting address */
+               rm_out = (uint8_t *)rte_pktmbuf_append(m_out, (e >> 3));
+               if (rm_out == NULL) {
+                       op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+                       rte_bbdev_log(ERR,
+                                       "Too little space in output mbuf");
+                       return;
+               }
+               /* rte_bbdev_op_data.offset can be different than the offset
+                * of the appended bytes
+                */
+               rm_out = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
+
+               /* index of current code block */
+               rm_req.r = cb_idx;
+               /* total number of code block */
+               rm_req.C = c;
+               /* For DL - 1, UL - 0 */
+               rm_req.direction = 1;
+               /* According to 3ggp 36.212 Spec 5.1.4.1.2 section Nsoft, KMIMO
+                * and MDL_HARQ are used for Ncb calculation. As Ncb is already
+                * known we can adjust those parameters
+                */
+               rm_req.Nsoft = ncb * rm_req.C;
+               rm_req.KMIMO = 1;
+               rm_req.MDL_HARQ = 1;
+               /* According to 3ggp 36.212 Spec 5.1.4.1.2 section Nl, Qm and G
+                * are used for E calculation. As E is already known we can
+                * adjust those parameters
+                */
+               rm_req.NL = e;
+               rm_req.Qm = 1;
+               rm_req.G = rm_req.NL * rm_req.Qm * rm_req.C;
+
+               rm_req.rvidx = enc->rv_index;
+               rm_req.Kidx = k_idx - 1;
+               rm_req.nLen = k + 4;
+               rm_req.tin0 = out0;
+               rm_req.tin1 = out1;
+               rm_req.tin2 = out2;
+               rm_resp.output = rm_out;
+               rm_resp.OutputLen = (e >> 3);
+               if (enc->op_flags & RTE_BBDEV_TURBO_RV_INDEX_BYPASS)
+                       rm_req.bypass_rvidx = 1;
+               else
+                       rm_req.bypass_rvidx = 0;
+
+               if (bblib_rate_match_dl(&rm_req, &rm_resp) != 0) {
+                       op->status |= 1 << RTE_BBDEV_DRV_ERROR;
+                       rte_bbdev_log(ERR, "Rate matching failed");
+                       return;
+               }
+               enc->output.length += rm_resp.OutputLen;
+       } else {
+               /* Rate matching is bypassed */
+
+               /* Completing last byte of out0 (where 4 tail bits are stored)
+                * by moving first 4 bits from out1
+                */
+               tmp_out = (uint8_t *) --out1;
+               *tmp_out = *tmp_out | ((*(tmp_out + 1) & 0xF0) >> 4);
+               tmp_out++;
+               /* Shifting out1 data by 4 bits to the left */
+               for (m = 0; m < k >> 3; ++m) {
+                       uint8_t *first = tmp_out;
+                       uint8_t second = *(tmp_out + 1);
+                       *first = (*first << 4) | ((second & 0xF0) >> 4);
+                       tmp_out++;
+               }
+               /* Shifting out2 data by 8 bits to the left */
+               for (m = 0; m < (k >> 3) + 1; ++m) {
+                       *tmp_out = *(tmp_out + 1);
+                       tmp_out++;
+               }
+               *tmp_out = 0;
+
+               /* copy shifted output to turbo_enc entity */
+               out0 = (uint8_t *)rte_pktmbuf_append(m_out,
+                               (k >> 3) * 3 + 2);
+               if (out0 == NULL) {
+                       op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+                       rte_bbdev_log(ERR,
+                                       "Too little space in output mbuf");
+                       return;
+               }
+               enc->output.length += (k >> 3) * 3 + 2;
+               /* rte_bbdev_op_data.offset can be different than the
+                * offset of the appended bytes
+                */
+               out0 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
+                               out_offset);
+               rte_memcpy(out0, q->enc_out, (k >> 3) * 3 + 2);
+       }
+}
+
+static inline void
+enqueue_enc_one_op(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op)
+{
+       uint8_t c, r, crc24_bits = 0;
+       uint16_t k, ncb;
+       uint32_t e;
+       struct rte_bbdev_op_turbo_enc *enc = &op->turbo_enc;
+       uint16_t in_offset = enc->input.offset;
+       uint16_t out_offset = enc->output.offset;
+       struct rte_mbuf *m_in = enc->input.data;
+       struct rte_mbuf *m_out = enc->output.data;
+       uint16_t total_left = enc->input.length;
+
+       /* Clear op status */
+       op->status = 0;
+
+       if (total_left > MAX_TB_SIZE >> 3) {
+               rte_bbdev_log(ERR, "TB size (%u) is too big, max: %d",
+                               total_left, MAX_TB_SIZE);
+               op->status = 1 << RTE_BBDEV_DATA_ERROR;
+               return;
+       }
+
+       if (m_in == NULL || m_out == NULL) {
+               rte_bbdev_log(ERR, "Invalid mbuf pointer");
+               op->status = 1 << RTE_BBDEV_DATA_ERROR;
+               return;
+       }
+
+       if ((enc->op_flags & RTE_BBDEV_TURBO_CRC_24B_ATTACH) ||
+               (enc->op_flags & RTE_BBDEV_TURBO_CRC_24A_ATTACH))
+               crc24_bits = 24;
+
+       if (enc->code_block_mode == 0) { /* For Transport Block mode */
+               c = enc->tb_params.c;
+               r = enc->tb_params.r;
+       } else {/* For Code Block mode */
+               c = 1;
+               r = 0;
+       }
+
+       while (total_left > 0 && r < c) {
+               if (enc->code_block_mode == 0) {
+                       k = (r < enc->tb_params.c_neg) ?
+                               enc->tb_params.k_neg : enc->tb_params.k_pos;
+                       ncb = (r < enc->tb_params.c_neg) ?
+                               enc->tb_params.ncb_neg : enc->tb_params.ncb_pos;
+                       e = (r < enc->tb_params.cab) ?
+                               enc->tb_params.ea : enc->tb_params.eb;
+               } else {
+                       k = enc->cb_params.k;
+                       ncb = enc->cb_params.ncb;
+                       e = enc->cb_params.e;
+               }
+
+               process_enc_cb(q, op, r, c, k, ncb, e, m_in,
+                               m_out, in_offset, out_offset, total_left);
+               /* Update total_left */
+               total_left -= (k - crc24_bits) >> 3;
+               /* Update offsets for next CBs (if exist) */
+               in_offset += (k - crc24_bits) >> 3;
+               if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH)
+                       out_offset += e >> 3;
+               else
+                       out_offset += (k >> 3) * 3 + 2;
+               r++;
+       }
+
+       /* check if all input data was processed */
+       if (total_left != 0) {
+               op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log(ERR,
+                               "Mismatch between mbuf length and included CBs sizes");
+       }
+}
+
+static inline uint16_t
+enqueue_enc_all_ops(struct turbo_sw_queue *q, struct rte_bbdev_enc_op **ops,
+               uint16_t nb_ops)
+{
+       uint16_t i;
+
+       for (i = 0; i < nb_ops; ++i)
+               enqueue_enc_one_op(q, ops[i]);
+
+       return rte_ring_enqueue_burst(q->processed_pkts, (void **)ops, nb_ops,
+                       NULL);
+}
+
+/* Remove the padding bytes from a cyclic buffer.
+ * The input buffer is a data stream wk as described in 3GPP TS 36.212 section
+ * 5.1.4.1.2 starting from w0 and with length Ncb bytes.
+ * The output buffer is a data stream wk with pruned padding bytes. It's length
+ * is 3*D bytes and the order of non-padding bytes is preserved.
+ */
+static inline void
+remove_nulls_from_circular_buf(const uint8_t *in, uint8_t *out, uint16_t k,
+               uint16_t ncb)
+{
+       uint32_t in_idx, out_idx, c_idx;
+       const uint32_t d = k + 4;
+       const uint32_t kw = (ncb / 3);
+       const uint32_t nd = kw - d;
+       const uint32_t r_subblock = kw / C_SUBBLOCK;
+       /* Inter-column permutation pattern */
+       const uint32_t P[C_SUBBLOCK] = {0, 16, 8, 24, 4, 20, 12, 28, 2, 18, 10,
+                       26, 6, 22, 14, 30, 1, 17, 9, 25, 5, 21, 13, 29, 3, 19,
+                       11, 27, 7, 23, 15, 31};
+       in_idx = 0;
+       out_idx = 0;
+
+       /* The padding bytes are at the first Nd positions in the first row. */
+       for (c_idx = 0; in_idx < kw; in_idx += r_subblock, ++c_idx) {
+               if (P[c_idx] < nd) {
+                       rte_memcpy(&out[out_idx], &in[in_idx + 1],
+                                       r_subblock - 1);
+                       out_idx += r_subblock - 1;
+               } else {
+                       rte_memcpy(&out[out_idx], &in[in_idx], r_subblock);
+                       out_idx += r_subblock;
+               }
+       }
+
+       /* First and second parity bits sub-blocks are interlaced. */
+       for (c_idx = 0; in_idx < ncb - 2 * r_subblock;
+                       in_idx += 2 * r_subblock, ++c_idx) {
+               uint32_t second_block_c_idx = P[c_idx];
+               uint32_t third_block_c_idx = P[c_idx] + 1;
+
+               if (second_block_c_idx < nd && third_block_c_idx < nd) {
+                       rte_memcpy(&out[out_idx], &in[in_idx + 2],
+                                       2 * r_subblock - 2);
+                       out_idx += 2 * r_subblock - 2;
+               } else if (second_block_c_idx >= nd &&
+                               third_block_c_idx >= nd) {
+                       rte_memcpy(&out[out_idx], &in[in_idx], 2 * r_subblock);
+                       out_idx += 2 * r_subblock;
+               } else if (second_block_c_idx < nd) {
+                       out[out_idx++] = in[in_idx];
+                       rte_memcpy(&out[out_idx], &in[in_idx + 2],
+                                       2 * r_subblock - 2);
+                       out_idx += 2 * r_subblock - 2;
+               } else {
+                       rte_memcpy(&out[out_idx], &in[in_idx + 1],
+                                       2 * r_subblock - 1);
+                       out_idx += 2 * r_subblock - 1;
+               }
+       }
+
+       /* Last interlaced row is different - its last byte is the only padding
+        * byte. We can have from 2 up to 26 padding bytes (Nd) per sub-block.
+        * After interlacing the 1st and 2nd parity sub-blocks we can have 0, 1
+        * or 2 padding bytes each time we make a step of 2 * R_SUBBLOCK bytes
+        * (moving to another column). 2nd parity sub-block uses the same
+        * inter-column permutation pattern as the systematic and 1st parity
+        * sub-blocks but it adds '1' to the resulting index and calculates the
+        * modulus of the result and Kw. Last column is mapped to itself (id 31)
+        * so the first byte taken from the 2nd parity sub-block will be the
+        * 32nd (31+1) byte, then 64th etc. (step is C_SUBBLOCK == 32) and the
+        * last byte will be the first byte from the sub-block:
+        * (32 + 32 * (R_SUBBLOCK-1)) % Kw == Kw % Kw == 0. Nd can't  be smaller
+        * than 2 so we know that bytes with ids 0 and 1 must be the padding
+        * bytes. The bytes from the 1st parity sub-block are the bytes from the
+        * 31st column - Nd can't be greater than 26 so we are sure that there
+        * are no padding bytes in 31st column.
+        */
+       rte_memcpy(&out[out_idx], &in[in_idx], 2 * r_subblock - 1);
+}
+
+static inline void
+move_padding_bytes(const uint8_t *in, uint8_t *out, uint16_t k,
+               uint16_t ncb)
+{
+       uint16_t d = k + 4;
+       uint16_t kpi = ncb / 3;
+       uint16_t nd = kpi - d;
+
+       rte_memcpy(&out[nd], in, d);
+       rte_memcpy(&out[nd + kpi + 64], &in[kpi], d);
+       rte_memcpy(&out[nd + 2 * (kpi + 64)], &in[2 * kpi], d);
+}
+
+static inline void
+process_dec_cb(struct turbo_sw_queue *q, struct rte_bbdev_dec_op *op,
+               uint8_t c, uint16_t k, uint16_t kw, struct rte_mbuf *m_in,
+               struct rte_mbuf *m_out, uint16_t in_offset, uint16_t out_offset,
+               bool check_crc_24b, uint16_t total_left)
+{
+       int ret;
+       int32_t k_idx;
+       int32_t iter_cnt;
+       uint8_t *in, *out, *adapter_input;
+       int32_t ncb, ncb_without_null;
+       struct bblib_turbo_adapter_ul_response adapter_resp;
+       struct bblib_turbo_adapter_ul_request adapter_req;
+       struct bblib_turbo_decoder_request turbo_req;
+       struct bblib_turbo_decoder_response turbo_resp;
+       struct rte_bbdev_op_turbo_dec *dec = &op->turbo_dec;
+
+       k_idx = compute_idx(k);
+
+       ret = is_dec_input_valid(k_idx, kw, total_left);
+       if (ret != 0) {
+               op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+               return;
+       }
+
+       in = rte_pktmbuf_mtod_offset(m_in, uint8_t *, in_offset);
+       ncb = kw;
+       ncb_without_null = (k + 4) * 3;
+
+       if (check_bit(dec->op_flags, RTE_BBDEV_TURBO_SUBBLOCK_DEINTERLEAVE)) {
+               struct bblib_deinterleave_ul_request deint_req;
+               struct bblib_deinterleave_ul_response deint_resp;
+
+               /* SW decoder accepts only a circular buffer without NULL bytes
+                * so the input needs to be converted.
+                */
+               remove_nulls_from_circular_buf(in, q->deint_input, k, ncb);
+
+               deint_req.pharqbuffer = q->deint_input;
+               deint_req.ncb = ncb_without_null;
+               deint_resp.pinteleavebuffer = q->deint_output;
+               bblib_deinterleave_ul(&deint_req, &deint_resp);
+       } else
+               move_padding_bytes(in, q->deint_output, k, ncb);
+
+       adapter_input = q->deint_output;
+
+       if (dec->op_flags & RTE_BBDEV_TURBO_POS_LLR_1_BIT_IN)
+               adapter_req.isinverted = 1;
+       else if (dec->op_flags & RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN)
+               adapter_req.isinverted = 0;
+       else {
+               op->status |= 1 << RTE_BBDEV_DRV_ERROR;
+               rte_bbdev_log(ERR, "LLR format wasn't specified");
+               return;
+       }
+
+       adapter_req.ncb = ncb_without_null;
+       adapter_req.pinteleavebuffer = adapter_input;
+       adapter_resp.pharqout = q->adapter_output;
+       bblib_turbo_adapter_ul(&adapter_req, &adapter_resp);
+
+       out = (uint8_t *)rte_pktmbuf_append(m_out, (k >> 3));
+       if (out == NULL) {
+               op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log(ERR, "Too little space in output mbuf");
+               return;
+       }
+       /* rte_bbdev_op_data.offset can be different than the offset of the
+        * appended bytes
+        */
+       out = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
+       if (check_crc_24b)
+               turbo_req.c = c + 1;
+       else
+               turbo_req.c = c;
+       turbo_req.input = (int8_t *)q->adapter_output;
+       turbo_req.k = k;
+       turbo_req.k_idx = k_idx;
+       turbo_req.max_iter_num = dec->iter_max;
+       turbo_resp.ag_buf = q->ag;
+       turbo_resp.cb_buf = q->code_block;
+       turbo_resp.output = out;
+       iter_cnt = bblib_turbo_decoder(&turbo_req, &turbo_resp);
+       dec->hard_output.length += (k >> 3);
+
+       if (iter_cnt > 0) {
+               /* Temporary solution for returned iter_count from SDK */
+               iter_cnt = (iter_cnt - 1) / 2;
+               dec->iter_count = RTE_MAX(iter_cnt, dec->iter_count);
+       } else {
+               op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log(ERR, "Turbo Decoder failed");
+               return;
+       }
+}
+
+static inline void
+enqueue_dec_one_op(struct turbo_sw_queue *q, struct rte_bbdev_dec_op *op)
+{
+       uint8_t c, r = 0;
+       uint16_t kw, k = 0;
+       struct rte_bbdev_op_turbo_dec *dec = &op->turbo_dec;
+       struct rte_mbuf *m_in = dec->input.data;
+       struct rte_mbuf *m_out = dec->hard_output.data;
+       uint16_t in_offset = dec->input.offset;
+       uint16_t total_left = dec->input.length;
+       uint16_t out_offset = dec->hard_output.offset;
+
+       /* Clear op status */
+       op->status = 0;
+
+       if (m_in == NULL || m_out == NULL) {
+               rte_bbdev_log(ERR, "Invalid mbuf pointer");
+               op->status = 1 << RTE_BBDEV_DATA_ERROR;
+               return;
+       }
+
+       if (dec->code_block_mode == 0) { /* For Transport Block mode */
+               c = dec->tb_params.c;
+       } else { /* For Code Block mode */
+               k = dec->cb_params.k;
+               c = 1;
+       }
+
+       while (total_left > 0) {
+               if (dec->code_block_mode == 0)
+                       k = (r < dec->tb_params.c_neg) ?
+                               dec->tb_params.k_neg : dec->tb_params.k_pos;
+
+               /* Calculates circular buffer size (Kw).
+                * According to 3gpp 36.212 section 5.1.4.2
+                *   Kw = 3 * Kpi,
+                * where:
+                *   Kpi = nCol * nRow
+                * where nCol is 32 and nRow can be calculated from:
+                *   D =< nCol * nRow
+                * where D is the size of each output from turbo encoder block
+                * (k + 4).
+                */
+               kw = RTE_ALIGN_CEIL(k + 4, C_SUBBLOCK) * 3;
+
+               process_dec_cb(q, op, c, k, kw, m_in, m_out, in_offset,
+                               out_offset, check_bit(dec->op_flags,
+                               RTE_BBDEV_TURBO_CRC_TYPE_24B), total_left);
+               /* As a result of decoding we get Code Block with included
+                * decoded CRC24 at the end of Code Block. Type of CRC24 is
+                * specified by flag.
+                */
+
+               /* Update total_left */
+               total_left -= kw;
+               /* Update offsets for next CBs (if exist) */
+               in_offset += kw;
+               out_offset += (k >> 3);
+               r++;
+       }
+       if (total_left != 0) {
+               op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+               rte_bbdev_log(ERR,
+                               "Mismatch between mbuf length and included Circular buffer sizes");
+       }
+}
+
+static inline uint16_t
+enqueue_dec_all_ops(struct turbo_sw_queue *q, struct rte_bbdev_dec_op **ops,
+               uint16_t nb_ops)
+{
+       uint16_t i;
+
+       for (i = 0; i < nb_ops; ++i)
+               enqueue_dec_one_op(q, ops[i]);
+
+       return rte_ring_enqueue_burst(q->processed_pkts, (void **)ops, nb_ops,
+                       NULL);
+}
+
+/* Enqueue burst */
+static uint16_t
+enqueue_enc_ops(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_enc_op **ops, uint16_t nb_ops)
+{
+       void *queue = q_data->queue_private;
+       struct turbo_sw_queue *q = queue;
+       uint16_t nb_enqueued = 0;
+
+       nb_enqueued = enqueue_enc_all_ops(q, ops, nb_ops);
+
+       q_data->queue_stats.enqueue_err_count += nb_ops - nb_enqueued;
+       q_data->queue_stats.enqueued_count += nb_enqueued;
+
+       return nb_enqueued;
+}
+
+/* Enqueue burst */
+static uint16_t
+enqueue_dec_ops(struct rte_bbdev_queue_data *q_data,
+                struct rte_bbdev_dec_op **ops, uint16_t nb_ops)
+{
+       void *queue = q_data->queue_private;
+       struct turbo_sw_queue *q = queue;
+       uint16_t nb_enqueued = 0;
+
+       nb_enqueued = enqueue_dec_all_ops(q, ops, nb_ops);
+
+       q_data->queue_stats.enqueue_err_count += nb_ops - nb_enqueued;
+       q_data->queue_stats.enqueued_count += nb_enqueued;
+
+       return nb_enqueued;
+}
+
+/* Dequeue decode burst */
+static uint16_t
+dequeue_dec_ops(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_dec_op **ops, uint16_t nb_ops)
+{
+       struct turbo_sw_queue *q = q_data->queue_private;
+       uint16_t nb_dequeued = rte_ring_dequeue_burst(q->processed_pkts,
+                       (void **)ops, nb_ops, NULL);
+       q_data->queue_stats.dequeued_count += nb_dequeued;
+
+       return nb_dequeued;
+}
+
+/* Dequeue encode burst */
+static uint16_t
+dequeue_enc_ops(struct rte_bbdev_queue_data *q_data,
+               struct rte_bbdev_enc_op **ops, uint16_t nb_ops)
+{
+       struct turbo_sw_queue *q = q_data->queue_private;
+       uint16_t nb_dequeued = rte_ring_dequeue_burst(q->processed_pkts,
+                       (void **)ops, nb_ops, NULL);
+       q_data->queue_stats.dequeued_count += nb_dequeued;
+
+       return nb_dequeued;
+}
+
+/* Parse 16bit integer from string argument */
+static inline int
+parse_u16_arg(const char *key, const char *value, void *extra_args)
+{
+       uint16_t *u16 = extra_args;
+       unsigned int long result;
+
+       if ((value == NULL) || (extra_args == NULL))
+               return -EINVAL;
+       errno = 0;
+       result = strtoul(value, NULL, 0);
+       if ((result >= (1 << 16)) || (errno != 0)) {
+               rte_bbdev_log(ERR, "Invalid value %lu for %s", result, key);
+               return -ERANGE;
+       }
+       *u16 = (uint16_t)result;
+       return 0;
+}
+
+/* Parse parameters used to create device */
+static int
+parse_turbo_sw_params(struct turbo_sw_params *params, const char *input_args)
+{
+       struct rte_kvargs *kvlist = NULL;
+       int ret = 0;
+
+       if (params == NULL)
+               return -EINVAL;
+       if (input_args) {
+               kvlist = rte_kvargs_parse(input_args, turbo_sw_valid_params);
+               if (kvlist == NULL)
+                       return -EFAULT;
+
+               ret = rte_kvargs_process(kvlist, turbo_sw_valid_params[0],
+                                       &parse_u16_arg, &params->queues_num);
+               if (ret < 0)
+                       goto exit;
+
+               ret = rte_kvargs_process(kvlist, turbo_sw_valid_params[1],
+                                       &parse_u16_arg, &params->socket_id);
+               if (ret < 0)
+                       goto exit;
+
+               if (params->socket_id >= RTE_MAX_NUMA_NODES) {
+                       rte_bbdev_log(ERR, "Invalid socket, must be < %u",
+                                       RTE_MAX_NUMA_NODES);
+                       goto exit;
+               }
+       }
+
+exit:
+       if (kvlist)
+               rte_kvargs_free(kvlist);
+       return ret;
+}
+
+/* Create device */
+static int
+turbo_sw_bbdev_create(struct rte_vdev_device *vdev,
+               struct turbo_sw_params *init_params)
+{
+       struct rte_bbdev *bbdev;
+       const char *name = rte_vdev_device_name(vdev);
+
+       bbdev = rte_bbdev_allocate(name);
+       if (bbdev == NULL)
+               return -ENODEV;
+
+       bbdev->data->dev_private = rte_zmalloc_socket(name,
+                       sizeof(struct bbdev_private), RTE_CACHE_LINE_SIZE,
+                       init_params->socket_id);
+       if (bbdev->data->dev_private == NULL) {
+               rte_bbdev_release(bbdev);
+               return -ENOMEM;
+       }
+
+       bbdev->dev_ops = &pmd_ops;
+       bbdev->device = &vdev->device;
+       bbdev->data->socket_id = init_params->socket_id;
+       bbdev->intr_handle = NULL;
+
+       /* register rx/tx burst functions for data path */
+       bbdev->dequeue_enc_ops = dequeue_enc_ops;
+       bbdev->dequeue_dec_ops = dequeue_dec_ops;
+       bbdev->enqueue_enc_ops = enqueue_enc_ops;
+       bbdev->enqueue_dec_ops = enqueue_dec_ops;
+       ((struct bbdev_private *) bbdev->data->dev_private)->max_nb_queues =
+                       init_params->queues_num;
+
+       return 0;
+}
+
+/* Initialise device */
+static int
+turbo_sw_bbdev_probe(struct rte_vdev_device *vdev)
+{
+       struct turbo_sw_params init_params = {
+               rte_socket_id(),
+               RTE_BBDEV_DEFAULT_MAX_NB_QUEUES
+       };
+       const char *name;
+       const char *input_args;
+
+       if (vdev == NULL)
+               return -EINVAL;
+
+       name = rte_vdev_device_name(vdev);
+       if (name == NULL)
+               return -EINVAL;
+       input_args = rte_vdev_device_args(vdev);
+       parse_turbo_sw_params(&init_params, input_args);
+
+       rte_bbdev_log_debug(
+                       "Initialising %s on NUMA node %d with max queues: %d\n",
+                       name, init_params.socket_id, init_params.queues_num);
+
+       return turbo_sw_bbdev_create(vdev, &init_params);
+}
+
+/* Uninitialise device */
+static int
+turbo_sw_bbdev_remove(struct rte_vdev_device *vdev)
+{
+       struct rte_bbdev *bbdev;
+       const char *name;
+
+       if (vdev == NULL)
+               return -EINVAL;
+
+       name = rte_vdev_device_name(vdev);
+       if (name == NULL)
+               return -EINVAL;
+
+       bbdev = rte_bbdev_get_named_dev(name);
+       if (bbdev == NULL)
+               return -EINVAL;
+
+       rte_free(bbdev->data->dev_private);
+
+       return rte_bbdev_release(bbdev);
+}
+
+static struct rte_vdev_driver bbdev_turbo_sw_pmd_drv = {
+       .probe = turbo_sw_bbdev_probe,
+       .remove = turbo_sw_bbdev_remove
+};
+
+RTE_PMD_REGISTER_VDEV(DRIVER_NAME, bbdev_turbo_sw_pmd_drv);
+RTE_PMD_REGISTER_PARAM_STRING(DRIVER_NAME,
+       TURBO_SW_MAX_NB_QUEUES_ARG"=<int> "
+       TURBO_SW_SOCKET_ID_ARG"=<int>");
+
+int bbdev_logtype;
+RTE_INIT(null_bbdev_init_log);
+static void
+null_bbdev_init_log(void)
+{
+       bbdev_logtype = rte_log_register("pmd.bbdev.turbo_sw");
+       if (bbdev_logtype >= 0)
+               rte_log_set_level(bbdev_logtype, RTE_LOG_NOTICE);
+}
diff --git a/drivers/bbdev/turbo_sw/rte_pmd_bbdev_turbo_sw_version.map b/drivers/bbdev/turbo_sw/rte_pmd_bbdev_turbo_sw_version.map
new file mode 100644 (file)
index 0000000..58b9427
--- /dev/null
@@ -0,0 +1,3 @@
+DPDK_18.02 {
+       local: *;
+};
index d27f17e..2dcac8f 100644 (file)
@@ -166,6 +166,14 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_VMXNET3_PMD)    += -lrte_pmd_vmxnet3_uio
 
 ifeq ($(CONFIG_RTE_LIBRTE_BBDEV),y)
 _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_NULL)     += -lrte_pmd_bbdev_null
+
+# TURBO SOFTWARE PMD is dependent on the FLEXRAN library
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -lrte_pmd_bbdev_turbo_sw
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -L$(FLEXRAN_SDK)/lib_crc -lcrc
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -L$(FLEXRAN_SDK)/lib_turbo -lturbo
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -L$(FLEXRAN_SDK)/lib_rate_matching -lrate_matching
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -L$(FLEXRAN_SDK)/lib_common -lcommon
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_BBDEV_TURBO_SW) += -lirc -limf -lstdc++ -lipps
 endif # CONFIG_RTE_LIBRTE_BBDEV
 
 ifeq ($(CONFIG_RTE_LIBRTE_CRYPTODEV),y)