Add support for disabling and enabling RX and TX DMA rings in mlx5en(4).

This is useful for supporting setups similar to Netmap.

Sponsored by:	Mellanox Technologies
MFC after:	1 week

sys/dev/mlx5/mlx5_en/en.h

@@ -408,6 +408,8 @@ struct mlx5e_params {
   m(+1, u64 tx_completion_fact_max, "tx_completion_fact_max", "Maximum completion event ratio") \
   m(+1, u64 hw_lro, "hw_lro", "set to enable hw_lro") \
   m(+1, u64 cqe_zipping, "cqe_zipping", "0 : CQE zipping disabled") \
+  m(+1, u64 modify_tx_dma, "modify_tx_dma", "0: Enable TX 1: Disable TX") \
+  m(+1, u64 modify_rx_dma, "modify_rx_dma", "0: Enable RX 1: Disable RX") \
   m(+1, u64 diag_pci_enable, "diag_pci_enable", "0: Disabled 1: Enabled") \
   m(+1, u64 diag_general_enable, "diag_general_enable", "0: Disabled 1: Enabled")
 
@@ -837,5 +839,8 @@ int	mlx5e_enable_sq(struct mlx5e_sq *, struct mlx5e_sq_param *, int tis_num);
 int	mlx5e_modify_sq(struct mlx5e_sq *, int curr_state, int next_state);
 void	mlx5e_disable_sq(struct mlx5e_sq *);
 void	mlx5e_drain_sq(struct mlx5e_sq *);
+void	mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value);
+void	mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value);
+void	mlx5e_resume_sq(struct mlx5e_sq *sq);
 
 #endif					/* _MLX5_EN_H_ */

sys/dev/mlx5/mlx5_en/mlx5_en_ethtool.c

@@ -377,6 +377,32 @@ mlx5e_ethtool_handler(SYSCTL_HANDLER_ARGS)
 			mlx5e_open_locked(priv->ifp);
 		break;
 
+	case MLX5_PARAM_OFFSET(modify_tx_dma):
+		/* check if network interface is opened */
+		if (was_opened) {
+			priv->params_ethtool.modify_tx_dma =
+			    priv->params_ethtool.modify_tx_dma ? 1 : 0;
+			/* modify tx according to value */
+			mlx5e_modify_tx_dma(priv, value != 0);
+		} else {
+			/* if closed force enable tx */
+			priv->params_ethtool.modify_tx_dma = 0;
+		}
+		break;
+
+	case MLX5_PARAM_OFFSET(modify_rx_dma):
+		/* check if network interface is opened */
+		if (was_opened) {
+			priv->params_ethtool.modify_rx_dma =
+			    priv->params_ethtool.modify_rx_dma ? 1 : 0;
+			/* modify rx according to value */
+			mlx5e_modify_rx_dma(priv, value != 0);
+		} else {
+			/* if closed force enable rx */
+			priv->params_ethtool.modify_rx_dma = 0;
+		}
+		break;
+
 	case MLX5_PARAM_OFFSET(diag_pci_enable):
 		priv->params_ethtool.diag_pci_enable =
 		    priv->params_ethtool.diag_pci_enable ? 1 : 0;

sys/dev/mlx5/mlx5_en/mlx5_en_main.c

@@ -2927,6 +2927,164 @@ sysctl_firmware(SYSCTL_HANDLER_ARGS)
 	return (error);
 }
 
+static void
+mlx5e_disable_tx_dma(struct mlx5e_channel *ch)
+{
+	int i;
+
+	for (i = 0; i < ch->num_tc; i++)
+		mlx5e_drain_sq(&ch->sq[i]);
+}
+
+static void
+mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq)
+{
+
+	sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP);
+	sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8);
+	mlx5e_tx_notify_hw(sq, sq->doorbell.d32, 0);
+	sq->doorbell.d64 = 0;
+}
+
+void
+mlx5e_resume_sq(struct mlx5e_sq *sq)
+{
+	int err;
+
+	/* check if already enabled */
+	if (sq->stopped == 0)
+		return;
+
+	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR,
+	    MLX5_SQC_STATE_RST);
+	if (err != 0) {
+		if_printf(sq->ifp,
+		    "mlx5e_modify_sq() from ERR to RST failed: %d\n", err);
+	}
+
+	sq->cc = 0;
+	sq->pc = 0;
+
+	/* reset doorbell prior to moving from RST to RDY */
+	mlx5e_reset_sq_doorbell_record(sq);
+
+	err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST,
+	    MLX5_SQC_STATE_RDY);
+	if (err != 0) {
+		if_printf(sq->ifp,
+		    "mlx5e_modify_sq() from RST to RDY failed: %d\n", err);
+	}
+
+	mtx_lock(&sq->lock);
+	sq->cev_next_state = MLX5E_CEV_STATE_INITIAL;
+	sq->stopped = 0;
+	mtx_unlock(&sq->lock);
+
+}
+
+static void
+mlx5e_enable_tx_dma(struct mlx5e_channel *ch)
+{
+        int i;
+
+	for (i = 0; i < ch->num_tc; i++)
+		mlx5e_resume_sq(&ch->sq[i]);
+}
+
+static void
+mlx5e_disable_rx_dma(struct mlx5e_channel *ch)
+{
+	struct mlx5e_rq *rq = &ch->rq;
+	int err;
+
+	mtx_lock(&rq->mtx);
+	rq->enabled = 0;
+	callout_stop(&rq->watchdog);
+	mtx_unlock(&rq->mtx);
+
+	callout_drain(&rq->watchdog);
+
+	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR);
+	if (err != 0) {
+		if_printf(rq->ifp,
+		    "mlx5e_modify_rq() from RDY to RST failed: %d\n", err);
+	}
+
+	while (!mlx5_wq_ll_is_empty(&rq->wq)) {
+		msleep(1);
+		rq->cq.mcq.comp(&rq->cq.mcq);
+	}
+
+	/*
+	 * Transitioning into RST state will allow the FW to track less ERR state queues,
+	 * thus reducing the recv queue flushing time
+	 */
+	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST);
+	if (err != 0) {
+		if_printf(rq->ifp,
+		    "mlx5e_modify_rq() from ERR to RST failed: %d\n", err);
+	}
+}
+
+static void
+mlx5e_enable_rx_dma(struct mlx5e_channel *ch)
+{
+	struct mlx5e_rq *rq = &ch->rq;
+	int err;
+
+	rq->wq.wqe_ctr = 0;
+	mlx5_wq_ll_update_db_record(&rq->wq);
+	err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
+	if (err != 0) {
+		if_printf(rq->ifp,
+		    "mlx5e_modify_rq() from RST to RDY failed: %d\n", err);
+        }
+
+	rq->enabled = 1;
+
+	rq->cq.mcq.comp(&rq->cq.mcq);
+}
+
+void
+mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value)
+{
+	int i;
+
+	if (priv->channel == NULL)
+		return;
+
+	for (i = 0; i < priv->params.num_channels; i++) {
+
+		if (!priv->channel[i])
+			continue;
+
+		if (value)
+			mlx5e_disable_tx_dma(priv->channel[i]);
+		else
+			mlx5e_enable_tx_dma(priv->channel[i]);
+	}
+}
+
+void
+mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value)
+{
+	int i;
+
+	if (priv->channel == NULL)
+		return;
+
+	for (i = 0; i < priv->params.num_channels; i++) {
+
+		if (!priv->channel[i])
+			continue;
+
+		if (value)
+			mlx5e_disable_rx_dma(priv->channel[i]);
+		else
+			mlx5e_enable_rx_dma(priv->channel[i]);
+	}
+}
+
 static void
 mlx5e_add_hw_stats(struct mlx5e_priv *priv)
 {