+
+ return 0;
+}
+
+static nicvf_iova_addr_t
+rbdr_rte_mempool_get(void *dev, void *opaque)
+{
+ uint16_t qidx;
+ uintptr_t mbuf;
+ struct nicvf_rxq *rxq;
+ struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
+ struct nicvf *nic = (struct nicvf *)opaque;
+ uint16_t rx_start, rx_end;
+
+ /* Get queue ranges for this VF */
+ nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
+
+ for (qidx = rx_start; qidx <= rx_end; qidx++) {
+ rxq = eth_dev->data->rx_queues[qidx];
+ /* Maintain equal buffer count across all pools */
+ if (rxq->precharge_cnt >= rxq->qlen_mask)
+ continue;
+ rxq->precharge_cnt++;
+ mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
+ if (mbuf)
+ return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
+ }
+ return 0;
+}
+
+static int
+nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
+{
+ int ret;
+ uint16_t qidx, data_off;
+ uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
+ uint64_t mbuf_phys_off = 0;
+ struct nicvf_rxq *rxq;
+ struct rte_mbuf *mbuf;
+ uint16_t rx_start, rx_end;
+ uint16_t tx_start, tx_end;
+ int mask;
+
+ PMD_INIT_FUNC_TRACE();
+
+ /* Userspace process exited without proper shutdown in last run */
+ if (nicvf_qset_rbdr_active(nic, 0))
+ nicvf_vf_stop(dev, nic, false);
+
+ /* Get queue ranges for this VF */
+ nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+ /*
+ * Thunderx nicvf PMD can support more than one pool per port only when
+ * 1) Data payload size is same across all the pools in given port
+ * AND
+ * 2) All mbuffs in the pools are from the same hugepage
+ * AND
+ * 3) Mbuff metadata size is same across all the pools in given port
+ *
+ * This is to support existing application that uses multiple pool/port.
+ * But, the purpose of using multipool for QoS will not be addressed.
+ *
+ */
+
+ /* Validate mempool attributes */
+ for (qidx = rx_start; qidx <= rx_end; qidx++) {
+ rxq = dev->data->rx_queues[qidx];
+ rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
+ mbuf = rte_pktmbuf_alloc(rxq->pool);
+ if (mbuf == NULL) {
+ PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
+ "pool=%s",
+ nic->vf_id, qidx, rxq->pool->name);
+ return -ENOMEM;
+ }
+ data_off = nicvf_mbuff_meta_length(mbuf);
+ data_off += RTE_PKTMBUF_HEADROOM;
+ rte_pktmbuf_free(mbuf);
+
+ if (data_off % RTE_CACHE_LINE_SIZE) {
+ PMD_INIT_LOG(ERR, "%s: unaligned data_off=%d delta=%d",
+ rxq->pool->name, data_off,
+ data_off % RTE_CACHE_LINE_SIZE);
+ return -EINVAL;
+ }
+ rxq->mbuf_phys_off -= data_off;
+ rxq->mbuf_phys_off -= nic->skip_bytes;
+
+ if (mbuf_phys_off == 0)
+ mbuf_phys_off = rxq->mbuf_phys_off;
+ if (mbuf_phys_off != rxq->mbuf_phys_off) {
+ PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
+ PRIx64, rxq->pool->name, nic->vf_id,
+ mbuf_phys_off);
+ return -EINVAL;
+ }
+ }
+
+ /* Check the level of buffers in the pool */
+ total_rxq_desc = 0;
+ for (qidx = rx_start; qidx <= rx_end; qidx++) {
+ rxq = dev->data->rx_queues[qidx];
+ /* Count total numbers of rxq descs */
+ total_rxq_desc += rxq->qlen_mask + 1;
+ exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
+ exp_buffs *= dev->data->nb_rx_queues;
+ if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
+ PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
+ rxq->pool->name,
+ rte_mempool_avail_count(rxq->pool),
+ exp_buffs);
+ return -ENOENT;
+ }
+ }
+
+ /* Check RBDR desc overflow */
+ ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
+ if (ret == 0) {
+ PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
+ "VF%d", nic->vf_id);
+ return -ENOMEM;
+ }
+
+ /* Enable qset */
+ ret = nicvf_qset_config(nic);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
+ nic->vf_id);
+ return ret;
+ }
+
+ /* Allocate RBDR and RBDR ring desc */
+ nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
+ ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
+ "VF%d", nic->vf_id);
+ goto qset_reclaim;
+ }
+
+ /* Enable and configure RBDR registers */
+ ret = nicvf_qset_rbdr_config(nic, 0);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
+ nic->vf_id);
+ goto qset_rbdr_free;
+ }
+
+ /* Fill rte_mempool buffers in RBDR pool and precharge it */
+ ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
+ total_rxq_desc);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
+ nic->vf_id);
+ goto qset_rbdr_reclaim;
+ }
+
+ PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
+ nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
+
+ /* Configure VLAN Strip */
+ mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK |
+ ETH_VLAN_EXTEND_MASK;
+ ret = nicvf_vlan_offload_config(dev, mask);
+
+ /* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
+ * to the 64bit memory address.
+ * The alignment creates a hole in mbuf(between the end of headroom and
+ * packet data start). The new revision of the HW provides an option to
+ * disable the L3 alignment feature and make mbuf layout looks
+ * more like other NICs. For better application compatibility, disabling
+ * l3 alignment feature on the hardware revisions it supports
+ */
+ nicvf_apad_config(nic, false);
+
+ /* Get queue ranges for this VF */
+ nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
+ /* Configure TX queues */
+ for (qidx = tx_start; qidx <= tx_end; qidx++) {
+ ret = nicvf_vf_start_tx_queue(dev, nic,
+ qidx % MAX_SND_QUEUES_PER_QS);
+ if (ret)
+ goto start_txq_error;
+ }
+
+ /* Configure RX queues */
+ for (qidx = rx_start; qidx <= rx_end; qidx++) {
+ ret = nicvf_vf_start_rx_queue(dev, nic,
+ qidx % MAX_RCV_QUEUES_PER_QS);
+ if (ret)
+ goto start_rxq_error;
+ }
+
+ if (!nic->sqs_mode) {
+ /* Configure CPI algorithm */
+ ret = nicvf_configure_cpi(dev);
+ if (ret)
+ goto start_txq_error;
+
+ ret = nicvf_mbox_get_rss_size(nic);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to get rss table size");
+ goto qset_rss_error;
+ }
+
+ /* Configure RSS */
+ ret = nicvf_configure_rss(dev);
+ if (ret)
+ goto qset_rss_error;
+ }
+
+ /* Done; Let PF make the BGX's RX and TX switches to ON position */
+ nicvf_mbox_cfg_done(nic);
+ return 0;
+
+qset_rss_error:
+ nicvf_rss_term(nic);
+start_rxq_error:
+ for (qidx = rx_start; qidx <= rx_end; qidx++)
+ nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
+start_txq_error:
+ for (qidx = tx_start; qidx <= tx_end; qidx++)
+ nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
+qset_rbdr_reclaim:
+ nicvf_qset_rbdr_reclaim(nic, 0);
+ nicvf_rbdr_release_mbufs(dev, nic);
+qset_rbdr_free:
+ if (nic->rbdr) {
+ rte_free(nic->rbdr);
+ nic->rbdr = NULL;
+ }
+qset_reclaim:
+ nicvf_qset_reclaim(nic);
+ return ret;
+}
+
+static int
+nicvf_dev_start(struct rte_eth_dev *dev)
+{
+ uint16_t qidx;
+ int ret;
+ size_t i;
+ struct nicvf *nic = nicvf_pmd_priv(dev);
+ struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
+ uint16_t mtu;
+ uint32_t buffsz = 0, rbdrsz = 0;
+ struct rte_pktmbuf_pool_private *mbp_priv;
+ struct nicvf_rxq *rxq;
+
+ PMD_INIT_FUNC_TRACE();
+
+ /* This function must be called for a primary device */
+ assert_primary(nic);
+
+ /* Validate RBDR buff size */
+ for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+ rxq = dev->data->rx_queues[qidx];
+ mbp_priv = rte_mempool_get_priv(rxq->pool);
+ buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
+ if (buffsz % 128) {
+ PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
+ return -EINVAL;
+ }
+ if (rbdrsz == 0)
+ rbdrsz = buffsz;
+ if (rbdrsz != buffsz) {
+ PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
+ qidx, rbdrsz, buffsz);
+ return -EINVAL;
+ }
+ }
+
+ /* Configure loopback */
+ ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
+ return ret;
+ }
+
+ /* Reset all statistics counters attached to this port */
+ ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
+ if (ret) {
+ PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
+ return ret;
+ }
+
+ /* Setup scatter mode if needed by jumbo */
+ if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
+ 2 * VLAN_TAG_SIZE > buffsz)
+ dev->data->scattered_rx = 1;
+ if ((rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER) != 0)
+ dev->data->scattered_rx = 1;
+
+ /* Setup MTU based on max_rx_pkt_len or default */
+ mtu = dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME ?
+ dev->data->dev_conf.rxmode.max_rx_pkt_len
+ - RTE_ETHER_HDR_LEN : RTE_ETHER_MTU;
+
+ if (nicvf_dev_set_mtu(dev, mtu)) {
+ PMD_INIT_LOG(ERR, "Failed to set default mtu size");
+ return -EBUSY;
+ }
+
+ ret = nicvf_vf_start(dev, nic, rbdrsz);
+ if (ret != 0)
+ return ret;
+
+ for (i = 0; i < nic->sqs_count; i++) {
+ assert(nic->snicvf[i]);
+
+ ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
+ if (ret != 0)
+ return ret;
+ }
+
+ /* Configure callbacks based on offloads */
+ nicvf_set_tx_function(dev);
+ nicvf_set_rx_function(dev);
+
+ return 0;
+}
+
+static void
+nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
+{
+ size_t i;
+ int ret;
+ struct nicvf *nic = nicvf_pmd_priv(dev);
+
+ PMD_INIT_FUNC_TRACE();
+ dev->data->dev_started = 0;
+
+ /* Teardown secondary vf first */
+ for (i = 0; i < nic->sqs_count; i++) {
+ if (!nic->snicvf[i])
+ continue;
+
+ nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
+ }
+
+ /* Stop the primary VF now */
+ nicvf_vf_stop(dev, nic, cleanup);
+
+ /* Disable loopback */
+ ret = nicvf_loopback_config(nic, 0);
+ if (ret)
+ PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
+
+ /* Reclaim CPI configuration */
+ ret = nicvf_mbox_config_cpi(nic, 0);
+ if (ret)
+ PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
+}
+
+static int
+nicvf_dev_stop(struct rte_eth_dev *dev)
+{
+ PMD_INIT_FUNC_TRACE();
+
+ nicvf_dev_stop_cleanup(dev, false);
+
+ return 0;
+}
+
+static void
+nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
+{
+ int ret;
+ uint16_t qidx;
+ uint16_t tx_start, tx_end;
+ uint16_t rx_start, rx_end;
+
+ PMD_INIT_FUNC_TRACE();
+
+ if (cleanup) {
+ /* Let PF make the BGX's RX and TX switches to OFF position */
+ nicvf_mbox_shutdown(nic);
+ }
+
+ /* Disable VLAN Strip */
+ nicvf_vlan_hw_strip(nic, 0);
+
+ /* Get queue ranges for this VF */
+ nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
+ for (qidx = tx_start; qidx <= tx_end; qidx++)
+ nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
+
+ /* Get queue ranges for this VF */
+ nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+ /* Reclaim rq */
+ for (qidx = rx_start; qidx <= rx_end; qidx++)
+ nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
+
+ /* Reclaim RBDR */
+ ret = nicvf_qset_rbdr_reclaim(nic, 0);
+ if (ret)
+ PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
+
+ /* Move all charged buffers in RBDR back to pool */
+ if (nic->rbdr != NULL)
+ nicvf_rbdr_release_mbufs(dev, nic);
+
+ /* Disable qset */
+ ret = nicvf_qset_reclaim(nic);
+ if (ret)
+ PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
+
+ /* Disable all interrupts */
+ nicvf_disable_all_interrupts(nic);
+
+ /* Free RBDR SW structure */
+ if (nic->rbdr) {
+ rte_free(nic->rbdr);
+ nic->rbdr = NULL;
+ }
+}
+
+static int
+nicvf_dev_close(struct rte_eth_dev *dev)
+{
+ size_t i;
+ struct nicvf *nic = nicvf_pmd_priv(dev);
+
+ PMD_INIT_FUNC_TRACE();
+ if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+ return 0;
+
+ nicvf_dev_stop_cleanup(dev, true);
+ nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
+
+ for (i = 0; i < nic->sqs_count; i++) {
+ if (!nic->snicvf[i])
+ continue;
+
+ nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
+ }
+
+ return 0;
+}
+
+static int
+nicvf_request_sqs(struct nicvf *nic)
+{
+ size_t i;
+
+ assert_primary(nic);
+ assert(nic->sqs_count > 0);
+ assert(nic->sqs_count <= MAX_SQS_PER_VF);
+
+ /* Set no of Rx/Tx queues in each of the SQsets */
+ for (i = 0; i < nic->sqs_count; i++) {
+ if (nicvf_svf_empty())
+ rte_panic("Cannot assign sufficient number of "
+ "secondary queues to primary VF%" PRIu8 "\n",
+ nic->vf_id);
+
+ nic->snicvf[i] = nicvf_svf_pop();
+ nic->snicvf[i]->sqs_id = i;
+ }
+
+ return nicvf_mbox_request_sqs(nic);