[RFC PATCH 14/29] net/qdma: add routine for Tx queue initialization

[Aman Kumar <aman.kumar@vvdntech.in>]

defined routines to handle tx queue related ops.
this patch add support to rte_eth_dev_tx_queue*
apis for this PMD.

Signed-off-by: Aman Kumar <aman.kumar@vvdntech.in>
---
 drivers/net/qdma/qdma.h        |   8 +
 drivers/net/qdma/qdma_common.c |  74 +++++++++
 drivers/net/qdma/qdma_devops.c | 270 +++++++++++++++++++++++++++++++--
 3 files changed, 343 insertions(+), 9 deletions(-)

diff --git a/drivers/net/qdma/qdma.h b/drivers/net/qdma/qdma.h
index 5992473b33..8515ebe60e 100644
--- a/drivers/net/qdma/qdma.h
+++ b/drivers/net/qdma/qdma.h
@@ -42,6 +42,7 @@
 #define MIN_RX_PIDX_UPDATE_THRESHOLD (1)
 #define MIN_TX_PIDX_UPDATE_THRESHOLD (1)
 #define DEFAULT_MM_CMPT_CNT_THRESHOLD	(2)
+#define QDMA_TXQ_PIDX_UPDATE_INTERVAL	(1000) /* 100 uSec */
 
 #define WB_TIMEOUT		(100000)
 #define RESET_TIMEOUT		(60000)
@@ -198,6 +199,7 @@ struct qdma_tx_queue {
 	uint16_t			tx_desc_pend;
 	uint16_t			nb_tx_desc; /* No of TX descriptors. */
 	rte_spinlock_t			pidx_update_lock;
+	struct qdma_q_pidx_reg_info	q_pidx_info;
 	uint64_t			offloads; /* Tx offloads */
 
 	uint8_t				st_mode:1;/* dma-mode: MM or ST */
@@ -297,17 +299,23 @@ struct qdma_pci_dev {
 };
 
 void qdma_dev_ops_init(struct rte_eth_dev *dev);
+void qdma_txq_pidx_update(void *arg);
 int qdma_pf_csr_read(struct rte_eth_dev *dev);
 
 uint8_t qmda_get_desc_sz_idx(enum rte_pmd_qdma_bypass_desc_len);
 
 int qdma_init_rx_queue(struct qdma_rx_queue *rxq);
+void qdma_reset_tx_queue(struct qdma_tx_queue *txq);
 void qdma_reset_rx_queue(struct qdma_rx_queue *rxq);
 
 void qdma_clr_rx_queue_ctxts(struct rte_eth_dev *dev, uint32_t qid,
 				uint32_t mode);
 void qdma_inv_rx_queue_ctxts(struct rte_eth_dev *dev, uint32_t qid,
 				uint32_t mode);
+void qdma_clr_tx_queue_ctxts(struct rte_eth_dev *dev, uint32_t qid,
+				uint32_t mode);
+void qdma_inv_tx_queue_ctxts(struct rte_eth_dev *dev, uint32_t qid,
+				uint32_t mode);
 int qdma_identify_bars(struct rte_eth_dev *dev);
 int qdma_get_hw_version(struct rte_eth_dev *dev);
 
diff --git a/drivers/net/qdma/qdma_common.c b/drivers/net/qdma/qdma_common.c
index d39e642008..2650438e47 100644
--- a/drivers/net/qdma/qdma_common.c
+++ b/drivers/net/qdma/qdma_common.c
@@ -160,6 +160,80 @@ int qdma_init_rx_queue(struct qdma_rx_queue *rxq)
 	return -ENOMEM;
 }
 
+/*
+ * Tx queue reset
+ */
+void qdma_reset_tx_queue(struct qdma_tx_queue *txq)
+{
+	uint32_t i;
+	uint32_t sz;
+
+	txq->tx_fl_tail = 0;
+	if (txq->st_mode) {  /* ST-mode */
+		sz = sizeof(struct qdma_ul_st_h2c_desc);
+		/* Zero out HW ring memory */
+		for (i = 0; i < (sz * (txq->nb_tx_desc)); i++)
+			((volatile char *)txq->tx_ring)[i] = 0;
+	} else {
+		sz = sizeof(struct qdma_ul_mm_desc);
+		/* Zero out HW ring memory */
+		for (i = 0; i < (sz * (txq->nb_tx_desc)); i++)
+			((volatile char *)txq->tx_ring)[i] = 0;
+	}
+
+	/* Initialize SW ring entries */
+	for (i = 0; i < txq->nb_tx_desc; i++)
+		txq->sw_ring[i] = NULL;
+}
+
+void qdma_inv_tx_queue_ctxts(struct rte_eth_dev *dev,
+			     uint32_t qid, uint32_t mode)
+{
+	struct qdma_pci_dev *qdma_dev = dev->data->dev_private;
+	struct qdma_descq_sw_ctxt q_sw_ctxt;
+	struct qdma_descq_hw_ctxt q_hw_ctxt;
+	struct qdma_descq_credit_ctxt q_credit_ctxt;
+	struct qdma_hw_access *hw_access = qdma_dev->hw_access;
+
+	hw_access->qdma_sw_ctx_conf(dev, 0, qid, &q_sw_ctxt,
+			QDMA_HW_ACCESS_INVALIDATE);
+	hw_access->qdma_hw_ctx_conf(dev, 0, qid, &q_hw_ctxt,
+			QDMA_HW_ACCESS_INVALIDATE);
+
+	if (mode) {  /* ST-mode */
+		hw_access->qdma_credit_ctx_conf(dev, 0, qid,
+			&q_credit_ctxt, QDMA_HW_ACCESS_INVALIDATE);
+	}
+}
+
+/**
+ * Clear Tx queue contexts
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   Nothing.
+ */
+void qdma_clr_tx_queue_ctxts(struct rte_eth_dev *dev,
+			     uint32_t qid, uint32_t mode)
+{
+	struct qdma_pci_dev *qdma_dev = dev->data->dev_private;
+	struct qdma_descq_sw_ctxt q_sw_ctxt;
+	struct qdma_descq_credit_ctxt q_credit_ctxt;
+	struct qdma_descq_hw_ctxt q_hw_ctxt;
+	struct qdma_hw_access *hw_access = qdma_dev->hw_access;
+
+	hw_access->qdma_sw_ctx_conf(dev, 0, qid, &q_sw_ctxt,
+			QDMA_HW_ACCESS_CLEAR);
+	hw_access->qdma_hw_ctx_conf(dev, 0, qid, &q_hw_ctxt,
+			QDMA_HW_ACCESS_CLEAR);
+	if (mode) {  /* ST-mode */
+		hw_access->qdma_credit_ctx_conf(dev, 0, qid,
+			&q_credit_ctxt, QDMA_HW_ACCESS_CLEAR);
+	}
+}
+
 /* Utility function to find index of an element in an array */
 int index_of_array(uint32_t *arr, uint32_t n, uint32_t element)
 {
diff --git a/drivers/net/qdma/qdma_devops.c b/drivers/net/qdma/qdma_devops.c
index fefbbda012..e411c0f1be 100644
--- a/drivers/net/qdma/qdma_devops.c
+++ b/drivers/net/qdma/qdma_devops.c
@@ -573,13 +573,196 @@ int qdma_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 			    uint16_t nb_tx_desc, unsigned int socket_id,
 			    const struct rte_eth_txconf *tx_conf)
 {
-	(void)dev;
-	(void)tx_queue_id;
-	(void)nb_tx_desc;
-	(void)socket_id;
-	(void)tx_conf;
+	struct qdma_pci_dev *qdma_dev = dev->data->dev_private;
+	struct qdma_tx_queue *txq = NULL;
+	struct qdma_ul_mm_desc *tx_ring_mm;
+	struct qdma_ul_st_h2c_desc *tx_ring_st;
+	uint32_t sz;
+	uint8_t  *tx_ring_bypass;
+	int err = 0;
+
+	PMD_DRV_LOG(INFO, "Configuring Tx queue id:%d with %d desc\n",
+		    tx_queue_id, nb_tx_desc);
+
+	if (!qdma_dev->is_vf) {
+		err = qdma_dev_increment_active_queue
+				(qdma_dev->dma_device_index,
+				qdma_dev->func_id,
+				QDMA_DEV_Q_TYPE_H2C);
+		if (err != QDMA_SUCCESS)
+			return -EINVAL;
+	}
+	if (!qdma_dev->init_q_range) {
+		if (!qdma_dev->is_vf) {
+			err = qdma_pf_csr_read(dev);
+			if (err < 0) {
+				PMD_DRV_LOG(ERR, "CSR read failed\n");
+				goto tx_setup_err;
+			}
+		}
+		qdma_dev->init_q_range = 1;
+	}
+	/* allocate rx queue data structure */
+	txq = rte_zmalloc_socket("QDMA_TxQ", sizeof(struct qdma_tx_queue),
+						RTE_CACHE_LINE_SIZE, socket_id);
+	if (txq == NULL) {
+		PMD_DRV_LOG(ERR, "Memory allocation failed for "
+				"Tx queue SW structure\n");
+		err = -ENOMEM;
+		goto tx_setup_err;
+	}
+
+	txq->st_mode = qdma_dev->q_info[tx_queue_id].queue_mode;
+	txq->en_bypass = (qdma_dev->q_info[tx_queue_id].tx_bypass_mode) ? 1 : 0;
+	txq->bypass_desc_sz = qdma_dev->q_info[tx_queue_id].tx_bypass_desc_sz;
+
+	txq->nb_tx_desc = (nb_tx_desc + 1);
+	txq->queue_id = tx_queue_id;
+	txq->dev = dev;
+	txq->port_id = dev->data->port_id;
+	txq->func_id = qdma_dev->func_id;
+	txq->num_queues = dev->data->nb_tx_queues;
+	txq->tx_deferred_start = tx_conf->tx_deferred_start;
+
+	txq->ringszidx = index_of_array(qdma_dev->g_ring_sz,
+					QDMA_NUM_RING_SIZES, txq->nb_tx_desc);
+	if (txq->ringszidx < 0) {
+		PMD_DRV_LOG(ERR, "Expected Ring size %d not found\n",
+				txq->nb_tx_desc);
+		err = -EINVAL;
+		goto tx_setup_err;
+	}
+
+	if (qdma_dev->ip_type == EQDMA_SOFT_IP &&
+			qdma_dev->vivado_rel >= QDMA_VIVADO_2020_2) {
+		if (qdma_dev->dev_cap.desc_eng_mode ==
+				QDMA_DESC_ENG_BYPASS_ONLY) {
+			PMD_DRV_LOG(ERR,
+				"Bypass only mode design "
+				"is not supported\n");
+			return -ENOTSUP;
+		}
+
+		if (txq->en_bypass &&
+				qdma_dev->dev_cap.desc_eng_mode ==
+				QDMA_DESC_ENG_INTERNAL_ONLY) {
+			PMD_DRV_LOG(ERR,
+				"Tx qid %d config in bypass "
+				"mode not supported on "
+				"internal only mode design\n",
+				tx_queue_id);
+			return -ENOTSUP;
+		}
+	}
+
+	/* Allocate memory for TX descriptor ring */
+	if (txq->st_mode) {
+		if (!qdma_dev->dev_cap.st_en) {
+			PMD_DRV_LOG(ERR, "Streaming mode not enabled "
+					"in the hardware\n");
+			err = -EINVAL;
+			goto tx_setup_err;
+		}
+
+		if (txq->en_bypass &&
+			txq->bypass_desc_sz != 0)
+			sz = (txq->nb_tx_desc) * (txq->bypass_desc_sz);
+		else
+			sz = (txq->nb_tx_desc) *
+					sizeof(struct qdma_ul_st_h2c_desc);
+		txq->tx_mz = qdma_zone_reserve(dev, "TxHwRn", tx_queue_id, sz,
+						socket_id);
+		if (!txq->tx_mz) {
+			PMD_DRV_LOG(ERR, "Couldn't reserve memory for "
+					"ST H2C ring of size %d\n", sz);
+			err = -ENOMEM;
+			goto tx_setup_err;
+		}
+
+		txq->tx_ring = txq->tx_mz->addr;
+		tx_ring_st = (struct qdma_ul_st_h2c_desc *)txq->tx_ring;
+
+		tx_ring_bypass = (uint8_t *)txq->tx_ring;
+		/* Write-back status structure */
+		if (txq->en_bypass &&
+			txq->bypass_desc_sz != 0)
+			txq->wb_status = (struct wb_status *)&
+					tx_ring_bypass[(txq->nb_tx_desc - 1) *
+					(txq->bypass_desc_sz)];
+		else
+			txq->wb_status = (struct wb_status *)&
+					tx_ring_st[txq->nb_tx_desc - 1];
+	} else {
+		if (!qdma_dev->dev_cap.mm_en) {
+			PMD_DRV_LOG(ERR, "Memory mapped mode not "
+					"enabled in the hardware\n");
+			err = -EINVAL;
+			goto tx_setup_err;
+		}
+
+		if (txq->en_bypass &&
+			txq->bypass_desc_sz != 0)
+			sz = (txq->nb_tx_desc) * (txq->bypass_desc_sz);
+		else
+			sz = (txq->nb_tx_desc) * sizeof(struct qdma_ul_mm_desc);
+		txq->tx_mz = qdma_zone_reserve(dev, "TxHwRn", tx_queue_id,
+						sz, socket_id);
+		if (!txq->tx_mz) {
+			PMD_DRV_LOG(ERR, "Couldn't reserve memory for "
+					"MM H2C ring of size %d\n", sz);
+			err = -ENOMEM;
+			goto tx_setup_err;
+		}
+
+		txq->tx_ring = txq->tx_mz->addr;
+		tx_ring_mm = (struct qdma_ul_mm_desc *)txq->tx_ring;
+
+		/* Write-back status structure */
+
+		tx_ring_bypass = (uint8_t *)txq->tx_ring;
+		if (txq->en_bypass &&
+			txq->bypass_desc_sz != 0)
+			txq->wb_status = (struct wb_status *)&
+				tx_ring_bypass[(txq->nb_tx_desc - 1) *
+				(txq->bypass_desc_sz)];
+		else
+			txq->wb_status = (struct wb_status *)&
+				tx_ring_mm[txq->nb_tx_desc - 1];
+	}
+
+	PMD_DRV_LOG(INFO, "Tx ring phys addr: 0x%lX, Tx Ring virt addr: 0x%lX",
+	    (uint64_t)txq->tx_mz->iova, (uint64_t)txq->tx_ring);
+
+	/* Allocate memory for TX software ring */
+	sz = txq->nb_tx_desc * sizeof(struct rte_mbuf *);
+	txq->sw_ring = rte_zmalloc_socket("TxSwRn", sz,
+				RTE_CACHE_LINE_SIZE, socket_id);
+	if (txq->sw_ring == NULL) {
+		PMD_DRV_LOG(ERR, "Memory allocation failed for "
+				 "Tx queue SW ring\n");
+		err = -ENOMEM;
+		goto tx_setup_err;
+	}
+
+	rte_spinlock_init(&txq->pidx_update_lock);
+	dev->data->tx_queues[tx_queue_id] = txq;
 
 	return 0;
+
+tx_setup_err:
+	PMD_DRV_LOG(ERR, " Tx queue setup failed");
+	if (!qdma_dev->is_vf)
+		qdma_dev_decrement_active_queue(qdma_dev->dma_device_index,
+						qdma_dev->func_id,
+						QDMA_DEV_Q_TYPE_H2C);
+	if (txq) {
+		if (txq->tx_mz)
+			rte_memzone_free(txq->tx_mz);
+		if (txq->sw_ring)
+			rte_free(txq->sw_ring);
+		rte_free(txq);
+	}
+	return err;
 }
 
 void qdma_dev_tx_queue_release(struct rte_eth_dev *dev, uint16_t q_id)
@@ -983,9 +1166,54 @@ int qdma_dev_configure(struct rte_eth_dev *dev)
 
 int qdma_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qid)
 {
-	(void)dev;
-	(void)qid;
+	struct qdma_pci_dev *qdma_dev = dev->data->dev_private;
+	struct qdma_tx_queue *txq;
+	uint32_t queue_base =  qdma_dev->queue_base;
+	int err, bypass_desc_sz_idx;
+	struct qdma_descq_sw_ctxt q_sw_ctxt;
+	struct qdma_hw_access *hw_access = qdma_dev->hw_access;
+
+	txq = (struct qdma_tx_queue *)dev->data->tx_queues[qid];
 
+	memset(&q_sw_ctxt, 0, sizeof(struct qdma_descq_sw_ctxt));
+
+	bypass_desc_sz_idx = qmda_get_desc_sz_idx(txq->bypass_desc_sz);
+
+	qdma_reset_tx_queue(txq);
+	qdma_clr_tx_queue_ctxts(dev, (qid + queue_base), txq->st_mode);
+
+	if (txq->st_mode) {
+		q_sw_ctxt.desc_sz = SW_DESC_CNTXT_H2C_STREAM_DMA;
+	} else {
+		q_sw_ctxt.desc_sz = SW_DESC_CNTXT_MEMORY_MAP_DMA;
+		q_sw_ctxt.is_mm = 1;
+	}
+	q_sw_ctxt.wbi_chk = 1;
+	q_sw_ctxt.wbi_intvl_en = 1;
+	q_sw_ctxt.fnc_id = txq->func_id;
+	q_sw_ctxt.qen = 1;
+	q_sw_ctxt.rngsz_idx = txq->ringszidx;
+	q_sw_ctxt.bypass = txq->en_bypass;
+	q_sw_ctxt.wbk_en = 1;
+	q_sw_ctxt.ring_bs_addr = (uint64_t)txq->tx_mz->iova;
+
+	if (txq->en_bypass &&
+		txq->bypass_desc_sz != 0)
+		q_sw_ctxt.desc_sz = bypass_desc_sz_idx;
+
+	/* Set SW Context */
+	err = hw_access->qdma_sw_ctx_conf(dev, 0,
+			(qid + queue_base), &q_sw_ctxt,
+			QDMA_HW_ACCESS_WRITE);
+	if (err < 0)
+		return qdma_dev->hw_access->qdma_get_error_code(err);
+
+	txq->q_pidx_info.pidx = 0;
+	hw_access->qdma_queue_pidx_update(dev, qdma_dev->is_vf,
+		qid, 0, &txq->q_pidx_info);
+
+	dev->data->tx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STARTED;
+	txq->status = RTE_ETH_QUEUE_STATE_STARTED;
 	return 0;
 }
 
@@ -1185,8 +1413,32 @@ int qdma_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qid)
 
 int qdma_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qid)
 {
-	(void)dev;
-	(void)qid;
+	struct qdma_pci_dev *qdma_dev = dev->data->dev_private;
+	uint32_t queue_base =  qdma_dev->queue_base;
+	struct qdma_tx_queue *txq;
+	int cnt = 0;
+	uint16_t count;
+
+	txq = (struct qdma_tx_queue *)dev->data->tx_queues[qid];
+
+	txq->status = RTE_ETH_QUEUE_STATE_STOPPED;
+	/* Wait for TXQ to send out all packets. */
+	while (txq->wb_status->cidx != txq->q_pidx_info.pidx) {
+		usleep(10);
+		if (cnt++ > 10000)
+			break;
+	}
+
+	qdma_inv_tx_queue_ctxts(dev, (qid + queue_base), txq->st_mode);
+
+	/* Relinquish pending mbufs */
+	for (count = 0; count < txq->nb_tx_desc - 1; count++) {
+		rte_pktmbuf_free(txq->sw_ring[count]);
+		txq->sw_ring[count] = NULL;
+	}
+	qdma_reset_tx_queue(txq);
+
+	dev->data->tx_queue_state[qid] = RTE_ETH_QUEUE_STATE_STOPPED;
 
 	return 0;
 }
-- 
2.36.1