net/bnxt: fix Rx checksum flags
[dpdk.git] / drivers / net / liquidio / lio_rxtx.c
index 14113e1..8d705bf 100644 (file)
@@ -1,37 +1,8 @@
-/*
- *   BSD LICENSE
- *
- *   Copyright(c) 2017 Cavium, Inc.. All rights reserved.
- *   All rights reserved.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Cavium, Inc. nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER(S) OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2017 Cavium, Inc
  */
 
-#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
 #include <rte_cycles.h>
 #include <rte_malloc.h>
 
@@ -41,6 +12,9 @@
 #include "lio_rxtx.h"
 
 #define LIO_MAX_SG 12
+/* Flush iq if available tx_desc fall below LIO_FLUSH_WM */
+#define LIO_FLUSH_WM(_iq) ((_iq)->nb_desc / 2)
+#define LIO_PKT_IN_DONE_CNT_MASK 0x00000000FFFFFFFFULL
 
 static void
 lio_droq_compute_max_packet_bufs(struct lio_droq *droq)
@@ -67,7 +41,7 @@ lio_droq_destroy_ring_buffers(struct lio_droq *droq)
 {
        uint32_t i;
 
-       for (i = 0; i < droq->max_count; i++) {
+       for (i = 0; i < droq->nb_desc; i++) {
                if (droq->recv_buf_list[i].buffer) {
                        rte_pktmbuf_free((struct rte_mbuf *)
                                         droq->recv_buf_list[i].buffer);
@@ -78,28 +52,6 @@ lio_droq_destroy_ring_buffers(struct lio_droq *droq)
        lio_droq_reset_indices(droq);
 }
 
-static void *
-lio_recv_buffer_alloc(struct lio_device *lio_dev, int q_no)
-{
-       struct lio_droq *droq = lio_dev->droq[q_no];
-       struct rte_mempool *mpool = droq->mpool;
-       struct rte_mbuf *m;
-
-       m = rte_pktmbuf_alloc(mpool);
-       if (m == NULL) {
-               lio_dev_err(lio_dev, "Cannot allocate\n");
-               return NULL;
-       }
-
-       rte_mbuf_refcnt_set(m, 1);
-       m->next = NULL;
-       m->data_off = RTE_PKTMBUF_HEADROOM;
-       m->nb_segs = 1;
-       m->pool = mpool;
-
-       return m;
-}
-
 static int
 lio_droq_setup_ring_buffers(struct lio_device *lio_dev,
                            struct lio_droq *droq)
@@ -108,10 +60,11 @@ lio_droq_setup_ring_buffers(struct lio_device *lio_dev,
        uint32_t i;
        void *buf;
 
-       for (i = 0; i < droq->max_count; i++) {
-               buf = lio_recv_buffer_alloc(lio_dev, droq->q_no);
+       for (i = 0; i < droq->nb_desc; i++) {
+               buf = rte_pktmbuf_alloc(droq->mpool);
                if (buf == NULL) {
                        lio_dev_err(lio_dev, "buffer alloc failed\n");
+                       droq->stats.rx_alloc_failure++;
                        lio_droq_destroy_ring_buffers(droq);
                        return -ENOMEM;
                }
@@ -182,7 +135,7 @@ lio_alloc_info_buffer(struct lio_device *lio_dev,
 {
        droq->info_mz = rte_eth_dma_zone_reserve(lio_dev->eth_dev,
                                                 "info_list", droq->q_no,
-                                                (droq->max_count *
+                                                (droq->nb_desc *
                                                        LIO_DROQ_INFO_SIZE),
                                                 RTE_CACHE_LINE_SIZE,
                                                 socket_id);
@@ -190,7 +143,7 @@ lio_alloc_info_buffer(struct lio_device *lio_dev,
        if (droq->info_mz == NULL)
                return NULL;
 
-       droq->info_list_dma = droq->info_mz->phys_addr;
+       droq->info_list_dma = droq->info_mz->iova;
        droq->info_alloc_size = droq->info_mz->len;
        droq->info_base_addr = (size_t)droq->info_mz->addr;
 
@@ -224,10 +177,10 @@ lio_init_droq(struct lio_device *lio_dev, uint32_t q_no,
 
        c_refill_threshold = LIO_OQ_REFILL_THRESHOLD_CFG(lio_dev);
 
-       droq->max_count = num_descs;
+       droq->nb_desc = num_descs;
        droq->buffer_size = desc_size;
 
-       desc_ring_size = droq->max_count * LIO_DROQ_DESC_SIZE;
+       desc_ring_size = droq->nb_desc * LIO_DROQ_DESC_SIZE;
        droq->desc_ring_mz = rte_eth_dma_zone_reserve(lio_dev->eth_dev,
                                                      "droq", q_no,
                                                      desc_ring_size,
@@ -240,13 +193,13 @@ lio_init_droq(struct lio_device *lio_dev, uint32_t q_no,
                return -1;
        }
 
-       droq->desc_ring_dma = droq->desc_ring_mz->phys_addr;
+       droq->desc_ring_dma = droq->desc_ring_mz->iova;
        droq->desc_ring = (struct lio_droq_desc *)droq->desc_ring_mz->addr;
 
        lio_dev_dbg(lio_dev, "droq[%d]: desc_ring: virt: 0x%p, dma: %lx\n",
                    q_no, droq->desc_ring, (unsigned long)droq->desc_ring_dma);
        lio_dev_dbg(lio_dev, "droq[%d]: num_desc: %d\n", q_no,
-                   droq->max_count);
+                   droq->nb_desc);
 
        droq->info_list = lio_alloc_info_buffer(lio_dev, droq, socket_id);
        if (droq->info_list == NULL) {
@@ -255,7 +208,7 @@ lio_init_droq(struct lio_device *lio_dev, uint32_t q_no,
        }
 
        droq->recv_buf_list = rte_zmalloc_socket("recv_buf_list",
-                                                (droq->max_count *
+                                                (droq->nb_desc *
                                                        LIO_DROQ_RECVBUF_SIZE),
                                                 RTE_CACHE_LINE_SIZE,
                                                 socket_id);
@@ -292,11 +245,6 @@ lio_setup_droq(struct lio_device *lio_dev, int oq_no, int num_descs,
 
        PMD_INIT_FUNC_TRACE();
 
-       if (lio_dev->droq[oq_no]) {
-               lio_dev_dbg(lio_dev, "Droq %d in use\n", oq_no);
-               return 0;
-       }
-
        /* Allocate the DS for the new droq. */
        droq = rte_zmalloc_socket("ethdev RX queue", sizeof(*droq),
                                  RTE_CACHE_LINE_SIZE, socket_id);
@@ -321,7 +269,7 @@ lio_setup_droq(struct lio_device *lio_dev, int oq_no, int num_descs,
        /* Send credit for octeon output queues. credits are always
         * sent after the output queue is enabled.
         */
-       rte_write32(lio_dev->droq[oq_no]->max_count,
+       rte_write32(lio_dev->droq[oq_no]->nb_desc,
                    lio_dev->droq[oq_no]->pkts_credit_reg);
        rte_wmb();
 
@@ -360,13 +308,13 @@ lio_droq_refill_pullup_descs(struct lio_droq *droq,
                        do {
                                droq->refill_idx = lio_incr_index(
                                                        droq->refill_idx, 1,
-                                                       droq->max_count);
+                                                       droq->nb_desc);
                                desc_refilled++;
                                droq->refill_count--;
                        } while (droq->recv_buf_list[droq->refill_idx].buffer);
                }
                refill_index = lio_incr_index(refill_index, 1,
-                                             droq->max_count);
+                                             droq->nb_desc);
        }       /* while */
 
        return desc_refilled;
@@ -374,7 +322,6 @@ lio_droq_refill_pullup_descs(struct lio_droq *droq,
 
 /* lio_droq_refill
  *
- * @param lio_dev      - pointer to the lio device structure
  * @param droq         - droq in which descriptors require new buffers.
  *
  * Description:
@@ -390,7 +337,7 @@ lio_droq_refill_pullup_descs(struct lio_droq *droq,
  * This routine is called with droq->lock held.
  */
 static uint32_t
-lio_droq_refill(struct lio_device *lio_dev, struct lio_droq *droq)
+lio_droq_refill(struct lio_droq *droq)
 {
        struct lio_droq_desc *desc_ring;
        uint32_t desc_refilled = 0;
@@ -398,17 +345,19 @@ lio_droq_refill(struct lio_device *lio_dev, struct lio_droq *droq)
 
        desc_ring = droq->desc_ring;
 
-       while (droq->refill_count && (desc_refilled < droq->max_count)) {
+       while (droq->refill_count && (desc_refilled < droq->nb_desc)) {
                /* If a valid buffer exists (happens if there is no dispatch),
                 * reuse the buffer, else allocate.
                 */
                if (droq->recv_buf_list[droq->refill_idx].buffer == NULL) {
-                       buf = lio_recv_buffer_alloc(lio_dev, droq->q_no);
+                       buf = rte_pktmbuf_alloc(droq->mpool);
                        /* If a buffer could not be allocated, no point in
                         * continuing
                         */
-                       if (buf == NULL)
+                       if (buf == NULL) {
+                               droq->stats.rx_alloc_failure++;
                                break;
+                       }
 
                        droq->recv_buf_list[droq->refill_idx].buffer = buf;
                }
@@ -419,7 +368,7 @@ lio_droq_refill(struct lio_device *lio_dev, struct lio_droq *droq)
                droq->info_list[droq->refill_idx].length = 0;
 
                droq->refill_idx = lio_incr_index(droq->refill_idx, 1,
-                                                 droq->max_count);
+                                                 droq->nb_desc);
                desc_refilled++;
                droq->refill_count--;
        }
@@ -466,7 +415,7 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
                buf_cnt = lio_droq_get_bufcount(droq->buffer_size,
                                                (uint32_t)info->length);
                droq->read_idx = lio_incr_index(droq->read_idx, buf_cnt,
-                                               droq->max_count);
+                                               droq->nb_desc);
                droq->refill_count += buf_cnt;
        } else {
                if (info->length <= droq->buffer_size) {
@@ -479,13 +428,10 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
                        droq->recv_buf_list[droq->read_idx].buffer = NULL;
                        droq->read_idx = lio_incr_index(
                                                droq->read_idx, 1,
-                                               droq->max_count);
+                                               droq->nb_desc);
                        droq->refill_count++;
 
                        if (likely(nicbuf != NULL)) {
-                               nicbuf->data_off = RTE_PKTMBUF_HEADROOM;
-                               nicbuf->nb_segs = 1;
-                               nicbuf->next = NULL;
                                /* We don't have a way to pass flags yet */
                                nicbuf->ol_flags = 0;
                                if (rh->r_dh.has_hash) {
@@ -539,9 +485,6 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
                                        if (!pkt_len)
                                                first_buf = nicbuf;
 
-                                       nicbuf->data_off = RTE_PKTMBUF_HEADROOM;
-                                       nicbuf->nb_segs = 1;
-                                       nicbuf->next = NULL;
                                        nicbuf->port = lio_dev->port_id;
                                        /* We don't have a way to pass
                                         * flags yet
@@ -579,7 +522,7 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
                                pkt_len += cpy_len;
                                droq->read_idx = lio_incr_index(
                                                        droq->read_idx,
-                                                       1, droq->max_count);
+                                                       1, droq->nb_desc);
                                droq->refill_count++;
 
                                /* Prefetch buffer pointers when on a
@@ -611,7 +554,7 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
        }
 
        if (droq->refill_count >= droq->refill_threshold) {
-               int desc_refilled = lio_droq_refill(lio_dev, droq);
+               int desc_refilled = lio_droq_refill(droq);
 
                /* Flush the droq descriptor data to memory to be sure
                 * that when we update the credits the data in memory is
@@ -626,6 +569,11 @@ lio_droq_fast_process_packet(struct lio_device *lio_dev,
        info->length = 0;
        info->rh.rh64 = 0;
 
+       droq->stats.pkts_received++;
+       droq->stats.rx_pkts_received += data_pkts;
+       droq->stats.rx_bytes_received += data_total_len;
+       droq->stats.bytes_received += total_len;
+
        return data_pkts;
 }
 
@@ -752,10 +700,10 @@ lio_init_instr_queue(struct lio_device *lio_dev,
                return -1;
        }
 
-       iq->base_addr_dma = iq->iq_mz->phys_addr;
+       iq->base_addr_dma = iq->iq_mz->iova;
        iq->base_addr = (uint8_t *)iq->iq_mz->addr;
 
-       iq->max_count = num_descs;
+       iq->nb_desc = num_descs;
 
        /* Initialize a list to holds requests that have been posted to Octeon
         * but has yet to be fetched by octeon
@@ -774,7 +722,7 @@ lio_init_instr_queue(struct lio_device *lio_dev,
 
        lio_dev_dbg(lio_dev, "IQ[%d]: base: %p basedma: %lx count: %d\n",
                    iq_no, iq->base_addr, (unsigned long)iq->base_addr_dma,
-                   iq->max_count);
+                   iq->nb_desc);
 
        iq->lio_dev = lio_dev;
        iq->txpciq.txpciq64 = txpciq.txpciq64;
@@ -871,14 +819,6 @@ lio_setup_iq(struct lio_device *lio_dev, int q_index,
 {
        uint32_t iq_no = (uint32_t)txpciq.s.q_no;
 
-       if (lio_dev->instr_queue[iq_no]) {
-               lio_dev_dbg(lio_dev, "IQ is in use. Cannot create the IQ: %d again\n",
-                           iq_no);
-               lio_dev->instr_queue[iq_no]->txpciq.txpciq64 = txpciq.txpciq64;
-               lio_dev->instr_queue[iq_no]->app_ctx = app_ctx;
-               return 0;
-       }
-
        lio_dev->instr_queue[iq_no] = rte_zmalloc_socket("ethdev TX queue",
                                                sizeof(struct lio_instr_queue),
                                                RTE_CACHE_LINE_SIZE, socket_id);
@@ -888,23 +828,49 @@ lio_setup_iq(struct lio_device *lio_dev, int q_index,
        lio_dev->instr_queue[iq_no]->q_index = q_index;
        lio_dev->instr_queue[iq_no]->app_ctx = app_ctx;
 
-       if (lio_init_instr_queue(lio_dev, txpciq, num_descs, socket_id))
-               goto release_lio_iq;
+       if (lio_init_instr_queue(lio_dev, txpciq, num_descs, socket_id)) {
+               rte_free(lio_dev->instr_queue[iq_no]);
+               lio_dev->instr_queue[iq_no] = NULL;
+               return -1;
+       }
 
        lio_dev->num_iqs++;
-       if (lio_dev->fn_list.enable_io_queues(lio_dev))
-               goto delete_lio_iq;
 
        return 0;
+}
 
-delete_lio_iq:
-       lio_delete_instr_queue(lio_dev, iq_no);
-       lio_dev->num_iqs--;
-release_lio_iq:
-       rte_free(lio_dev->instr_queue[iq_no]);
-       lio_dev->instr_queue[iq_no] = NULL;
+int
+lio_wait_for_instr_fetch(struct lio_device *lio_dev)
+{
+       int pending, instr_cnt;
+       int i, retry = 1000;
 
-       return -1;
+       do {
+               instr_cnt = 0;
+
+               for (i = 0; i < LIO_MAX_INSTR_QUEUES(lio_dev); i++) {
+                       if (!(lio_dev->io_qmask.iq & (1ULL << i)))
+                               continue;
+
+                       if (lio_dev->instr_queue[i] == NULL)
+                               break;
+
+                       pending = rte_atomic64_read(
+                           &lio_dev->instr_queue[i]->instr_pending);
+                       if (pending)
+                               lio_flush_iq(lio_dev, lio_dev->instr_queue[i]);
+
+                       instr_cnt += pending;
+               }
+
+               if (instr_cnt == 0)
+                       break;
+
+               rte_delay_ms(1);
+
+       } while (retry-- && instr_cnt);
+
+       return instr_cnt;
 }
 
 static inline void
@@ -941,14 +907,14 @@ post_command2(struct lio_instr_queue *iq, uint8_t *cmd)
         * position if queue gets full before Octeon could fetch any instr.
         */
        if (rte_atomic64_read(&iq->instr_pending) >=
-                       (int32_t)(iq->max_count - 1)) {
+                       (int32_t)(iq->nb_desc - 1)) {
                st.status = LIO_IQ_SEND_FAILED;
                st.index = -1;
                return st;
        }
 
        if (rte_atomic64_read(&iq->instr_pending) >=
-                       (int32_t)(iq->max_count - 2))
+                       (int32_t)(iq->nb_desc - 2))
                st.status = LIO_IQ_SEND_STOP;
 
        copy_cmd_into_iq(iq, cmd);
@@ -956,7 +922,7 @@ post_command2(struct lio_instr_queue *iq, uint8_t *cmd)
        /* "index" is returned, host_write_index is modified. */
        st.index = iq->host_write_index;
        iq->host_write_index = lio_incr_index(iq->host_write_index, 1,
-                                             iq->max_count);
+                                             iq->nb_desc);
        iq->fill_cnt++;
 
        /* Flush the command into memory. We need to be sure the data is in
@@ -977,9 +943,151 @@ lio_add_to_request_list(struct lio_instr_queue *iq,
        iq->request_list[idx].reqtype = reqtype;
 }
 
+static inline void
+lio_free_netsgbuf(void *buf)
+{
+       struct lio_buf_free_info *finfo = buf;
+       struct lio_device *lio_dev = finfo->lio_dev;
+       struct rte_mbuf *m = finfo->mbuf;
+       struct lio_gather *g = finfo->g;
+       uint8_t iq = finfo->iq_no;
+
+       /* This will take care of multiple segments also */
+       rte_pktmbuf_free(m);
+
+       rte_spinlock_lock(&lio_dev->glist_lock[iq]);
+       STAILQ_INSERT_TAIL(&lio_dev->glist_head[iq], &g->list, entries);
+       rte_spinlock_unlock(&lio_dev->glist_lock[iq]);
+       rte_free(finfo);
+}
+
+/* Can only run in process context */
+static int
+lio_process_iq_request_list(struct lio_device *lio_dev,
+                           struct lio_instr_queue *iq)
+{
+       struct octeon_instr_irh *irh = NULL;
+       uint32_t old = iq->flush_index;
+       struct lio_soft_command *sc;
+       uint32_t inst_count = 0;
+       int reqtype;
+       void *buf;
+
+       while (old != iq->lio_read_index) {
+               reqtype = iq->request_list[old].reqtype;
+               buf     = iq->request_list[old].buf;
+
+               if (reqtype == LIO_REQTYPE_NONE)
+                       goto skip_this;
+
+               switch (reqtype) {
+               case LIO_REQTYPE_NORESP_NET:
+                       rte_pktmbuf_free((struct rte_mbuf *)buf);
+                       break;
+               case LIO_REQTYPE_NORESP_NET_SG:
+                       lio_free_netsgbuf(buf);
+                       break;
+               case LIO_REQTYPE_SOFT_COMMAND:
+                       sc = buf;
+                       irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
+                       if (irh->rflag) {
+                               /* We're expecting a response from Octeon.
+                                * It's up to lio_process_ordered_list() to
+                                * process sc. Add sc to the ordered soft
+                                * command response list because we expect
+                                * a response from Octeon.
+                                */
+                               rte_spinlock_lock(&lio_dev->response_list.lock);
+                               rte_atomic64_inc(
+                                   &lio_dev->response_list.pending_req_count);
+                               STAILQ_INSERT_TAIL(
+                                       &lio_dev->response_list.head,
+                                       &sc->node, entries);
+                               rte_spinlock_unlock(
+                                               &lio_dev->response_list.lock);
+                       } else {
+                               if (sc->callback) {
+                                       /* This callback must not sleep */
+                                       sc->callback(LIO_REQUEST_DONE,
+                                                    sc->callback_arg);
+                               }
+                       }
+                       break;
+               default:
+                       lio_dev_err(lio_dev,
+                                   "Unknown reqtype: %d buf: %p at idx %d\n",
+                                   reqtype, buf, old);
+               }
+
+               iq->request_list[old].buf = NULL;
+               iq->request_list[old].reqtype = 0;
+
+skip_this:
+               inst_count++;
+               old = lio_incr_index(old, 1, iq->nb_desc);
+       }
+
+       iq->flush_index = old;
+
+       return inst_count;
+}
+
+static void
+lio_update_read_index(struct lio_instr_queue *iq)
+{
+       uint32_t pkt_in_done = rte_read32(iq->inst_cnt_reg);
+       uint32_t last_done;
+
+       last_done = pkt_in_done - iq->pkt_in_done;
+       iq->pkt_in_done = pkt_in_done;
+
+       /* Add last_done and modulo with the IQ size to get new index */
+       iq->lio_read_index = (iq->lio_read_index +
+                       (uint32_t)(last_done & LIO_PKT_IN_DONE_CNT_MASK)) %
+                       iq->nb_desc;
+}
+
+int
+lio_flush_iq(struct lio_device *lio_dev, struct lio_instr_queue *iq)
+{
+       uint32_t tot_inst_processed = 0;
+       uint32_t inst_processed = 0;
+       int tx_done = 1;
+
+       if (rte_atomic64_test_and_set(&iq->iq_flush_running) == 0)
+               return tx_done;
+
+       rte_spinlock_lock(&iq->lock);
+
+       lio_update_read_index(iq);
+
+       do {
+               /* Process any outstanding IQ packets. */
+               if (iq->flush_index == iq->lio_read_index)
+                       break;
+
+               inst_processed = lio_process_iq_request_list(lio_dev, iq);
+
+               if (inst_processed) {
+                       rte_atomic64_sub(&iq->instr_pending, inst_processed);
+                       iq->stats.instr_processed += inst_processed;
+               }
+
+               tot_inst_processed += inst_processed;
+               inst_processed = 0;
+
+       } while (1);
+
+       rte_spinlock_unlock(&iq->lock);
+
+       rte_atomic64_clear(&iq->iq_flush_running);
+
+       return tx_done;
+}
+
 static int
 lio_send_command(struct lio_device *lio_dev, uint32_t iq_no, void *cmd,
-                void *buf, uint32_t datasize __rte_unused, uint32_t reqtype)
+                void *buf, uint32_t datasize, uint32_t reqtype)
 {
        struct lio_instr_queue *iq = lio_dev->instr_queue[iq_no];
        struct lio_iq_post_status st;
@@ -990,7 +1098,13 @@ lio_send_command(struct lio_device *lio_dev, uint32_t iq_no, void *cmd,
 
        if (st.status != LIO_IQ_SEND_FAILED) {
                lio_add_to_request_list(iq, st.index, buf, reqtype);
+               LIO_INCR_INSTRQUEUE_PKT_COUNT(lio_dev, iq_no, bytes_sent,
+                                             datasize);
+               LIO_INCR_INSTRQUEUE_PKT_COUNT(lio_dev, iq_no, instr_posted, 1);
+
                lio_ring_doorbell(lio_dev, iq);
+       } else {
+               LIO_INCR_INSTRQUEUE_PKT_COUNT(lio_dev, iq_no, instr_dropped, 1);
        }
 
        rte_spinlock_unlock(&iq->post_lock);
@@ -1134,7 +1248,7 @@ lio_alloc_soft_command(struct lio_device *lio_dev, uint32_t datasize,
        sc = rte_pktmbuf_mtod(m, struct lio_soft_command *);
        memset(sc, 0, LIO_SOFT_COMMAND_BUFFER_SIZE);
        sc->size = LIO_SOFT_COMMAND_BUFFER_SIZE;
-       sc->dma_addr = rte_mbuf_data_dma_addr(m);
+       sc->dma_addr = rte_mbuf_data_iova(m);
        sc->mbuf = m;
 
        dma_addr = sc->dma_addr;
@@ -1296,7 +1410,7 @@ list_delete_first_node(struct lio_stailq_head *head)
        return node;
 }
 
-static void
+void
 lio_delete_sglist(struct lio_instr_queue *txq)
 {
        struct lio_device *lio_dev = txq->lio_dev;
@@ -1385,6 +1499,117 @@ lio_delete_instruction_queue(struct lio_device *lio_dev, int iq_no)
        lio_dev->num_iqs--;
 }
 
+static inline uint32_t
+lio_iq_get_available(struct lio_device *lio_dev, uint32_t q_no)
+{
+       return ((lio_dev->instr_queue[q_no]->nb_desc - 1) -
+               (uint32_t)rte_atomic64_read(
+                               &lio_dev->instr_queue[q_no]->instr_pending));
+}
+
+static inline int
+lio_iq_is_full(struct lio_device *lio_dev, uint32_t q_no)
+{
+       return ((uint32_t)rte_atomic64_read(
+                               &lio_dev->instr_queue[q_no]->instr_pending) >=
+                               (lio_dev->instr_queue[q_no]->nb_desc - 2));
+}
+
+static int
+lio_dev_cleanup_iq(struct lio_device *lio_dev, int iq_no)
+{
+       struct lio_instr_queue *iq = lio_dev->instr_queue[iq_no];
+       uint32_t count = 10000;
+
+       while ((lio_iq_get_available(lio_dev, iq_no) < LIO_FLUSH_WM(iq)) &&
+                       --count)
+               lio_flush_iq(lio_dev, iq);
+
+       return count ? 0 : 1;
+}
+
+static void
+lio_ctrl_cmd_callback(uint32_t status __rte_unused, void *sc_ptr)
+{
+       struct lio_soft_command *sc = sc_ptr;
+       struct lio_dev_ctrl_cmd *ctrl_cmd;
+       struct lio_ctrl_pkt *ctrl_pkt;
+
+       ctrl_pkt = (struct lio_ctrl_pkt *)sc->ctxptr;
+       ctrl_cmd = ctrl_pkt->ctrl_cmd;
+       ctrl_cmd->cond = 1;
+
+       lio_free_soft_command(sc);
+}
+
+static inline struct lio_soft_command *
+lio_alloc_ctrl_pkt_sc(struct lio_device *lio_dev,
+                     struct lio_ctrl_pkt *ctrl_pkt)
+{
+       struct lio_soft_command *sc = NULL;
+       uint32_t uddsize, datasize;
+       uint32_t rdatasize;
+       uint8_t *data;
+
+       uddsize = (uint32_t)(ctrl_pkt->ncmd.s.more * 8);
+
+       datasize = OCTEON_CMD_SIZE + uddsize;
+       rdatasize = (ctrl_pkt->wait_time) ? 16 : 0;
+
+       sc = lio_alloc_soft_command(lio_dev, datasize,
+                                   rdatasize, sizeof(struct lio_ctrl_pkt));
+       if (sc == NULL)
+               return NULL;
+
+       rte_memcpy(sc->ctxptr, ctrl_pkt, sizeof(struct lio_ctrl_pkt));
+
+       data = (uint8_t *)sc->virtdptr;
+
+       rte_memcpy(data, &ctrl_pkt->ncmd, OCTEON_CMD_SIZE);
+
+       lio_swap_8B_data((uint64_t *)data, OCTEON_CMD_SIZE >> 3);
+
+       if (uddsize) {
+               /* Endian-Swap for UDD should have been done by caller. */
+               rte_memcpy(data + OCTEON_CMD_SIZE, ctrl_pkt->udd, uddsize);
+       }
+
+       sc->iq_no = (uint32_t)ctrl_pkt->iq_no;
+
+       lio_prepare_soft_command(lio_dev, sc,
+                                LIO_OPCODE, LIO_OPCODE_CMD,
+                                0, 0, 0);
+
+       sc->callback = lio_ctrl_cmd_callback;
+       sc->callback_arg = sc;
+       sc->wait_time = ctrl_pkt->wait_time;
+
+       return sc;
+}
+
+int
+lio_send_ctrl_pkt(struct lio_device *lio_dev, struct lio_ctrl_pkt *ctrl_pkt)
+{
+       struct lio_soft_command *sc = NULL;
+       int retval;
+
+       sc = lio_alloc_ctrl_pkt_sc(lio_dev, ctrl_pkt);
+       if (sc == NULL) {
+               lio_dev_err(lio_dev, "soft command allocation failed\n");
+               return -1;
+       }
+
+       retval = lio_send_soft_command(lio_dev, sc);
+       if (retval == LIO_IQ_SEND_FAILED) {
+               lio_free_soft_command(sc);
+               lio_dev_err(lio_dev, "Port: %d soft command: %d send failed status: %x\n",
+                           lio_dev->port_id, ctrl_pkt->ncmd.s.cmd, retval);
+               return -1;
+       }
+
+       return retval;
+}
+
 /** Send data packet to the device
  *  @param lio_dev - lio device pointer
  *  @param ndata   - control structure with queueing, and buffer information
@@ -1405,6 +1630,7 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
        struct lio_instr_queue *txq = tx_queue;
        union lio_cmd_setup cmdsetup;
        struct lio_device *lio_dev;
+       struct lio_iq_stats *stats;
        struct lio_data_pkt ndata;
        int i, processed = 0;
        struct rte_mbuf *m;
@@ -1414,13 +1640,16 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
 
        lio_dev = txq->lio_dev;
        iq_no = txq->txpciq.s.q_no;
+       stats = &lio_dev->instr_queue[iq_no]->stats;
 
-       if (!lio_dev->linfo.link.s.link_up) {
+       if (!lio_dev->intf_open || !lio_dev->linfo.link.s.link_up) {
                PMD_TX_LOG(lio_dev, ERR, "Transmit failed link_status : %d\n",
                           lio_dev->linfo.link.s.link_up);
                goto xmit_failed;
        }
 
+       lio_dev_cleanup_iq(lio_dev, iq_no);
+
        for (i = 0; i < nb_pkts; i++) {
                uint32_t pkt_len = 0;
 
@@ -1432,6 +1661,15 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
                ndata.buf = m;
 
                ndata.q_no = iq_no;
+               if (lio_iq_is_full(lio_dev, ndata.q_no)) {
+                       stats->tx_iq_busy++;
+                       if (lio_dev_cleanup_iq(lio_dev, iq_no)) {
+                               PMD_TX_LOG(lio_dev, ERR,
+                                          "Transmit failed iq:%d full\n",
+                                          ndata.q_no);
+                               break;
+                       }
+               }
 
                cmdsetup.cmd_setup64 = 0;
                cmdsetup.s.iq_no = iq_no;
@@ -1440,7 +1678,9 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
                if (m->ol_flags & PKT_TX_IP_CKSUM)
                        cmdsetup.s.ip_csum = 1;
 
-               if ((m->ol_flags & PKT_TX_TCP_CKSUM) ||
+               if (m->ol_flags & PKT_TX_OUTER_IP_CKSUM)
+                       cmdsetup.s.tnl_csum = 1;
+               else if ((m->ol_flags & PKT_TX_TCP_CKSUM) ||
                                (m->ol_flags & PKT_TX_UDP_CKSUM))
                        cmdsetup.s.transport_csum = 1;
 
@@ -1449,8 +1689,70 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
                        cmdsetup.s.u.datasize = pkt_len;
                        lio_prepare_pci_cmd(lio_dev, &ndata.cmd,
                                            &cmdsetup, tag);
-                       ndata.cmd.cmd3.dptr = rte_mbuf_data_dma_addr(m);
+                       ndata.cmd.cmd3.dptr = rte_mbuf_data_iova(m);
                        ndata.reqtype = LIO_REQTYPE_NORESP_NET;
+               } else {
+                       struct lio_buf_free_info *finfo;
+                       struct lio_gather *g;
+                       rte_iova_t phyaddr;
+                       int i, frags;
+
+                       finfo = (struct lio_buf_free_info *)rte_malloc(NULL,
+                                                       sizeof(*finfo), 0);
+                       if (finfo == NULL) {
+                               PMD_TX_LOG(lio_dev, ERR,
+                                          "free buffer alloc failed\n");
+                               goto xmit_failed;
+                       }
+
+                       rte_spinlock_lock(&lio_dev->glist_lock[iq_no]);
+                       g = (struct lio_gather *)list_delete_first_node(
+                                               &lio_dev->glist_head[iq_no]);
+                       rte_spinlock_unlock(&lio_dev->glist_lock[iq_no]);
+                       if (g == NULL) {
+                               PMD_TX_LOG(lio_dev, ERR,
+                                          "Transmit scatter gather: glist null!\n");
+                               goto xmit_failed;
+                       }
+
+                       cmdsetup.s.gather = 1;
+                       cmdsetup.s.u.gatherptrs = m->nb_segs;
+                       lio_prepare_pci_cmd(lio_dev, &ndata.cmd,
+                                           &cmdsetup, tag);
+
+                       memset(g->sg, 0, g->sg_size);
+                       g->sg[0].ptr[0] = rte_mbuf_data_iova(m);
+                       lio_add_sg_size(&g->sg[0], m->data_len, 0);
+                       pkt_len = m->data_len;
+                       finfo->mbuf = m;
+
+                       /* First seg taken care above */
+                       frags = m->nb_segs - 1;
+                       i = 1;
+                       m = m->next;
+                       while (frags--) {
+                               g->sg[(i >> 2)].ptr[(i & 3)] =
+                                               rte_mbuf_data_iova(m);
+                               lio_add_sg_size(&g->sg[(i >> 2)],
+                                               m->data_len, (i & 3));
+                               pkt_len += m->data_len;
+                               i++;
+                               m = m->next;
+                       }
+
+                       phyaddr = rte_mem_virt2iova(g->sg);
+                       if (phyaddr == RTE_BAD_IOVA) {
+                               PMD_TX_LOG(lio_dev, ERR, "bad phys addr\n");
+                               goto xmit_failed;
+                       }
+
+                       ndata.cmd.cmd3.dptr = phyaddr;
+                       ndata.reqtype = LIO_REQTYPE_NORESP_NET_SG;
+
+                       finfo->g = g;
+                       finfo->lio_dev = lio_dev;
+                       finfo->iq_no = (uint64_t)iq_no;
+                       ndata.buf = finfo;
                }
 
                ndata.datasize = pkt_len;
@@ -1462,14 +1764,43 @@ lio_dev_xmit_pkts(void *tx_queue, struct rte_mbuf **pkts, uint16_t nb_pkts)
                        break;
                }
 
-               if (unlikely(status == LIO_IQ_SEND_STOP))
+               if (unlikely(status == LIO_IQ_SEND_STOP)) {
                        PMD_TX_LOG(lio_dev, DEBUG, "iq full\n");
+                       /* create space as iq is full */
+                       lio_dev_cleanup_iq(lio_dev, iq_no);
+               }
 
+               stats->tx_done++;
+               stats->tx_tot_bytes += pkt_len;
                processed++;
        }
 
 xmit_failed:
+       stats->tx_dropped += (nb_pkts - processed);
 
        return processed;
 }
 
+void
+lio_dev_clear_queues(struct rte_eth_dev *eth_dev)
+{
+       struct lio_instr_queue *txq;
+       struct lio_droq *rxq;
+       uint16_t i;
+
+       for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
+               txq = eth_dev->data->tx_queues[i];
+               if (txq != NULL) {
+                       lio_dev_tx_queue_release(txq);
+                       eth_dev->data->tx_queues[i] = NULL;
+               }
+       }
+
+       for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
+               rxq = eth_dev->data->rx_queues[i];
+               if (rxq != NULL) {
+                       lio_dev_rx_queue_release(rxq);
+                       eth_dev->data->rx_queues[i] = NULL;
+               }
+       }
+}