[dpdk-dev] [PATCH v1 16/30] net/dpaa2: add taildrop support on frame count basis

[Sachin Saxena <sachin.saxena@nxp.com>]

From: Hemant Agrawal <hemant.agrawal@nxp.com>

The existing taildrop was based on queue data size.
This patch replaces it with frame count bases using
CGR methods of DPAA2 device.
The number of CGRs are limited. So,
- use per queue CGR based tail drop for as many as CGR
available.
- Remaining queues shall use the legacy byte based tail drop

Number of CGRs can be controlled by dpl file during dpni_create.

Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
 drivers/bus/fslmc/portal/dpaa2_hw_pvt.h |   5 +-
 drivers/net/dpaa2/dpaa2_ethdev.c        | 112 +++++++++++++++++++++---
 drivers/net/dpaa2/dpaa2_ethdev.h        |   6 +-
 3 files changed, 106 insertions(+), 17 deletions(-)

diff --git a/drivers/bus/fslmc/portal/dpaa2_hw_pvt.h b/drivers/bus/fslmc/portal/dpaa2_hw_pvt.h
index 4bb6b26c7..7f7e2fd78 100644
--- a/drivers/bus/fslmc/portal/dpaa2_hw_pvt.h
+++ b/drivers/bus/fslmc/portal/dpaa2_hw_pvt.h
@@ -145,10 +145,10 @@ struct dpaa2_queue {
 		struct rte_eth_dev_data *eth_data;
 		struct rte_cryptodev_data *crypto_data;
 	};
-	int32_t eventfd;	/*!< Event Fd of this queue */
 	uint32_t fqid;		/*!< Unique ID of this queue */
-	uint8_t tc_index;	/*!< traffic class identifier */
 	uint16_t flow_id;	/*!< To be used by DPAA2 frmework */
+	uint8_t tc_index;	/*!< traffic class identifier */
+	uint8_t cgid;		/*! < Congestion Group id for this queue */
 	uint64_t rx_pkts;
 	uint64_t tx_pkts;
 	uint64_t err_pkts;
@@ -157,6 +157,7 @@ struct dpaa2_queue {
 		struct qbman_result *cscn;
 	};
 	struct rte_event ev;
+	int32_t eventfd;	/*!< Event Fd of this queue */
 	dpaa2_queue_cb_dqrr_t *cb;
 	dpaa2_queue_cb_eqresp_free_t *cb_eqresp_free;
 	struct dpaa2_bp_info *bp_array;
diff --git a/drivers/net/dpaa2/dpaa2_ethdev.c b/drivers/net/dpaa2/dpaa2_ethdev.c
index f25cdfb3d..b30f37ecb 100644
--- a/drivers/net/dpaa2/dpaa2_ethdev.c
+++ b/drivers/net/dpaa2/dpaa2_ethdev.c
@@ -514,7 +514,7 @@ dpaa2_eth_dev_configure(struct rte_eth_dev *dev)
 static int
 dpaa2_dev_rx_queue_setup(struct rte_eth_dev *dev,
 			 uint16_t rx_queue_id,
-			 uint16_t nb_rx_desc __rte_unused,
+			 uint16_t nb_rx_desc,
 			 unsigned int socket_id __rte_unused,
 			 const struct rte_eth_rxconf *rx_conf __rte_unused,
 			 struct rte_mempool *mb_pool)
@@ -526,7 +526,7 @@ dpaa2_dev_rx_queue_setup(struct rte_eth_dev *dev,
 	uint8_t options = 0;
 	uint8_t flow_id;
 	uint32_t bpid;
-	int ret;
+	int i, ret;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -545,12 +545,28 @@ dpaa2_dev_rx_queue_setup(struct rte_eth_dev *dev,
 	dpaa2_q->bp_array = rte_dpaa2_bpid_info;
 
 	/*Get the flow id from given VQ id*/
-	flow_id = rx_queue_id % priv->nb_rx_queues;
+	flow_id = dpaa2_q->flow_id;
 	memset(&cfg, 0, sizeof(struct dpni_queue));
 
 	options = options | DPNI_QUEUE_OPT_USER_CTX;
 	cfg.user_context = (size_t)(dpaa2_q);
 
+	/* check if a private cgr available. */
+	for (i = 0; i < priv->max_cgs; i++) {
+		if (!priv->cgid_in_use[i]) {
+			priv->cgid_in_use[i] = 1;
+			break;
+		}
+	}
+
+	if (i < priv->max_cgs) {
+		options |= DPNI_QUEUE_OPT_SET_CGID;
+		cfg.cgid = i;
+		dpaa2_q->cgid = cfg.cgid;
+	} else {
+		dpaa2_q->cgid = 0xff;
+	}
+
 	/*if ls2088 or rev2 device, enable the stashing */
 
 	if ((dpaa2_svr_family & 0xffff0000) != SVR_LS2080A) {
@@ -579,15 +595,56 @@ dpaa2_dev_rx_queue_setup(struct rte_eth_dev *dev,
 		struct dpni_taildrop taildrop;
 
 		taildrop.enable = 1;
-		/*enabling per rx queue congestion control */
-		taildrop.threshold = CONG_THRESHOLD_RX_Q;
-		taildrop.units = DPNI_CONGESTION_UNIT_BYTES;
-		taildrop.oal = CONG_RX_OAL;
-		DPAA2_PMD_DEBUG("Enabling Early Drop on queue = %d",
-				rx_queue_id);
-		ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
+
+		/* Private CGR will use tail drop length as nb_rx_desc.
+		 * for rest cases we can use standard byte based tail drop.
+		 * There is no HW restriction, but number of CGRs are limited,
+		 * hence this restriction is placed.
+		 */
+		if (dpaa2_q->cgid != 0xff) {
+			/*enabling per rx queue congestion control */
+			taildrop.threshold = nb_rx_desc;
+			taildrop.units = DPNI_CONGESTION_UNIT_FRAMES;
+			taildrop.oal = 0;
+			DPAA2_PMD_DEBUG("Enabling CG Tail Drop on queue = %d",
+					rx_queue_id);
+			ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
+						DPNI_CP_CONGESTION_GROUP,
+						DPNI_QUEUE_RX,
+						dpaa2_q->tc_index,
+						flow_id, &taildrop);
+		} else {
+			/*enabling per rx queue congestion control */
+			taildrop.threshold = CONG_THRESHOLD_RX_BYTES_Q;
+			taildrop.units = DPNI_CONGESTION_UNIT_BYTES;
+			taildrop.oal = CONG_RX_OAL;
+			DPAA2_PMD_DEBUG("Enabling Byte based Drop on queue= %d",
+					rx_queue_id);
+			ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
+						DPNI_CP_QUEUE, DPNI_QUEUE_RX,
+						dpaa2_q->tc_index, flow_id,
+						&taildrop);
+		}
+		if (ret) {
+			DPAA2_PMD_ERR("Error in setting taildrop. err=(%d)",
+				      ret);
+			return -1;
+		}
+	} else { /* Disable tail Drop */
+		struct dpni_taildrop taildrop = {};
+		DPAA2_PMD_INFO("Tail drop is disabled on queue");
+
+		taildrop.enable = 0;
+		if (dpaa2_q->cgid != 0xff) {
+			ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
+					DPNI_CP_CONGESTION_GROUP, DPNI_QUEUE_RX,
+					dpaa2_q->tc_index,
+					flow_id, &taildrop);
+		} else {
+			ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
 					DPNI_CP_QUEUE, DPNI_QUEUE_RX,
 					dpaa2_q->tc_index, flow_id, &taildrop);
+		}
 		if (ret) {
 			DPAA2_PMD_ERR("Error in setting taildrop. err=(%d)",
 				      ret);
@@ -655,7 +712,7 @@ dpaa2_dev_tx_queue_setup(struct rte_eth_dev *dev,
 	dpaa2_q->tc_index = tc_id;
 
 	if (!(priv->flags & DPAA2_TX_CGR_OFF)) {
-		struct dpni_congestion_notification_cfg cong_notif_cfg;
+		struct dpni_congestion_notification_cfg cong_notif_cfg = {};
 
 		cong_notif_cfg.units = DPNI_CONGESTION_UNIT_FRAMES;
 		cong_notif_cfg.threshold_entry = CONG_ENTER_TX_THRESHOLD;
@@ -693,7 +750,28 @@ dpaa2_dev_tx_queue_setup(struct rte_eth_dev *dev,
 static void
 dpaa2_dev_rx_queue_release(void *q __rte_unused)
 {
+	struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)q;
+	struct dpaa2_dev_priv *priv = dpaa2_q->eth_data->dev_private;
+	struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+	uint8_t options = 0;
+	int ret;
+	struct dpni_queue cfg = {};
+
 	PMD_INIT_FUNC_TRACE();
+	if (dpaa2_q->cgid != 0xff) {
+		options = DPNI_QUEUE_OPT_CLEAR_CGID;
+		cfg.cgid = dpaa2_q->cgid;
+
+		ret = dpni_set_queue(dpni, CMD_PRI_LOW, priv->token,
+				     DPNI_QUEUE_RX,
+				     dpaa2_q->tc_index, dpaa2_q->flow_id,
+				     options, &cfg);
+		if (ret)
+			DPAA2_PMD_ERR("Unable to clear CGR from q=%u err=%d",
+					dpaa2_q->fqid, ret);
+		priv->cgid_in_use[dpaa2_q->cgid] = 0;
+		dpaa2_q->cgid = 0xff;
+	}
 }
 
 static void
@@ -2166,6 +2244,14 @@ dpaa2_dev_init(struct rte_eth_dev *eth_dev)
 	}
 
 	priv->num_rx_tc = attr.num_rx_tcs;
+	/* only if the custom CG is enabled */
+	if (attr.options & DPNI_OPT_CUSTOM_CG)
+		priv->max_cgs = attr.num_cgs;
+	else
+		priv->max_cgs = 0;
+
+	for (i = 0; i < priv->max_cgs; i++)
+		priv->cgid_in_use[i] = 0;
 
 	for (i = 0; i < attr.num_rx_tcs; i++)
 		priv->nb_rx_queues += attr.num_queues;
@@ -2173,9 +2259,9 @@ dpaa2_dev_init(struct rte_eth_dev *eth_dev)
 	/* Using number of TX queues as number of TX TCs */
 	priv->nb_tx_queues = attr.num_tx_tcs;
 
-	DPAA2_PMD_DEBUG("RX-TC= %d, nb_rx_queues= %d, nb_tx_queues=%d",
+	DPAA2_PMD_DEBUG("RX-TC= %d, rx_queues= %d, tx_queues=%d, max_cgs=%d",
 			priv->num_rx_tc, priv->nb_rx_queues,
-			priv->nb_tx_queues);
+			priv->nb_tx_queues, priv->max_cgs);
 
 	priv->hw = dpni_dev;
 	priv->hw_id = hw_id;
diff --git a/drivers/net/dpaa2/dpaa2_ethdev.h b/drivers/net/dpaa2/dpaa2_ethdev.h
index a991ccc1d..2f14a3525 100644
--- a/drivers/net/dpaa2/dpaa2_ethdev.h
+++ b/drivers/net/dpaa2/dpaa2_ethdev.h
@@ -37,9 +37,9 @@
 #define CONG_RETRY_COUNT 18000
 
 /* RX queue tail drop threshold
- * currently considering 32 KB packets
+ * currently considering 64 KB packets
  */
-#define CONG_THRESHOLD_RX_Q  (64 * 1024)
+#define CONG_THRESHOLD_RX_BYTES_Q  (64 * 1024)
 #define CONG_RX_OAL	128
 
 /* Size of the input SMMU mapped memory required by MC */
@@ -115,6 +115,8 @@ struct dpaa2_dev_priv {
 	uint8_t flags; /*dpaa2 config flags */
 	uint8_t en_ordered;
 	uint8_t en_loose_ordered;
+	uint8_t max_cgs;
+	uint8_t cgid_in_use[MAX_RX_QUEUES];
 
 	struct pattern_s {
 		uint8_t item_count;
-- 
2.17.1