[V3,3/3] i2c: tegra: Add DMA Support
diff mbox series

Message ID 1548475073-12408-3-git-send-email-skomatineni@nvidia.com
State Superseded
Headers show
Series
  • [V3,1/3] i2c: tegra: Sort all the include headers alphabetically
Related show

Commit Message

Sowjanya Komatineni Jan. 26, 2019, 3:57 a.m. UTC
This patch adds DMA support for Tegra I2C.

Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
transfer size of the max FIFO depth and DMA mode is used for
transfer size higher than max FIFO depth to save CPU overhead.

PIO mode needs full intervention of CPU to fill or empty FIFO's
and also need to service multiple data requests interrupt for the
same transaction adding overhead on CPU for large transfers.

DMA mode is helpful for Large transfers during downloading or
uploading FW over I2C to some external devices.

Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>

---
 [V3] : Updated without additional buffer allocation.
 [V2] : Updated based on V1 review feedback along with code cleanup for
	proper implementation of DMA.

 drivers/i2c/busses/i2c-tegra.c | 341 ++++++++++++++++++++++++++++++++++++++---
 1 file changed, 316 insertions(+), 25 deletions(-)

Comments

Dmitry Osipenko Jan. 26, 2019, 5:11 p.m. UTC | #1
26.01.2019 6:57, Sowjanya Komatineni пишет:
> This patch adds DMA support for Tegra I2C.
> 
> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
> transfer size of the max FIFO depth and DMA mode is used for
> transfer size higher than max FIFO depth to save CPU overhead.
> 
> PIO mode needs full intervention of CPU to fill or empty FIFO's
> and also need to service multiple data requests interrupt for the
> same transaction adding overhead on CPU for large transfers.
> 
> DMA mode is helpful for Large transfers during downloading or
> uploading FW over I2C to some external devices.
> 
> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> 
> ---
>  [V3] : Updated without additional buffer allocation.
>  [V2] : Updated based on V1 review feedback along with code cleanup for
> 	proper implementation of DMA.
> 
>  drivers/i2c/busses/i2c-tegra.c | 341 ++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 316 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
> index 3dcbc9960d9d..452358a77400 100644
> --- a/drivers/i2c/busses/i2c-tegra.c
> +++ b/drivers/i2c/busses/i2c-tegra.c
> @@ -8,6 +8,9 @@
>  
>  #include <linux/clk.h>
>  #include <linux/delay.h>
> +#include <linux/dmaengine.h>
> +#include <linux/dmapool.h>
> +#include <linux/dma-mapping.h>
>  #include <linux/err.h>
>  #include <linux/i2c.h>
>  #include <linux/init.h>
> @@ -45,6 +48,8 @@
>  #define I2C_FIFO_CONTROL_RX_FLUSH		BIT(0)
>  #define I2C_FIFO_CONTROL_TX_TRIG_SHIFT		5
>  #define I2C_FIFO_CONTROL_RX_TRIG_SHIFT		2
> +#define I2C_FIFO_CONTROL_TX_TRIG(x)		(((x) - 1) << 5)
> +#define I2C_FIFO_CONTROL_RX_TRIG(x)		(((x) - 1) << 2)
>  #define I2C_FIFO_STATUS				0x060
>  #define I2C_FIFO_STATUS_TX_MASK			0xF0
>  #define I2C_FIFO_STATUS_TX_SHIFT		4
> @@ -119,6 +124,16 @@
>  /* Packet header size in bytes */
>  #define I2C_PACKET_HEADER_SIZE			12
>  
> +#define DATA_DMA_DIR_TX				(1 << 0)
> +#define DATA_DMA_DIR_RX				(1 << 1)
> +
> +/*
> + * Upto I2C_PIO_MODE_MAX_LEN bytes, controller will use PIO mode,
> + * above this, controller will use DMA to fill FIFO.
> + * MAX PIO len is 20 bytes excluding packet header.
> + */
> +#define I2C_PIO_MODE_MAX_LEN			20
> +
>  /*
>   * msg_end_type: The bus control which need to be send at end of transfer.
>   * @MSG_END_STOP: Send stop pulse at end of transfer.
> @@ -179,6 +194,7 @@ struct tegra_i2c_hw_feature {
>   * @fast_clk: clock reference for fast clock of I2C controller
>   * @rst: reset control for the I2C controller
>   * @base: ioremapped registers cookie
> + * @phys_addr: Physical address of I2C base address to use for DMA configuration
>   * @cont_id: I2C controller ID, used for packet header
>   * @irq: IRQ number of transfer complete interrupt
>   * @irq_disabled: used to track whether or not the interrupt is enabled
> @@ -192,6 +208,14 @@ struct tegra_i2c_hw_feature {
>   * @clk_divisor_non_hs_mode: clock divider for non-high-speed modes
>   * @is_multimaster_mode: track if I2C controller is in multi-master mode
>   * @xfer_lock: lock to serialize transfer submission and processing
> + * @has_dma: indicated if controller supports DMA

AFAIK, all Terga's support DMA. Let's change to "@has_dma: can utilize DMA" for clarity.

> + * @tx_dma_chan: DMA transmit channel
> + * @rx_dma_chan: DMA receive channel
> + * @dma_phys: handle to DMA resources
> + * @dma_buf: pointer to allocated DMA buffer
> + * @dma_buf_size: DMA buffer size
> + * @is_curr_dma_xfer: indicates active DMA transfer
> + * @dma_complete: DMA completion notifier
>   */
>  struct tegra_i2c_dev {
>  	struct device *dev;
> @@ -201,6 +225,7 @@ struct tegra_i2c_dev {
>  	struct clk *fast_clk;
>  	struct reset_control *rst;
>  	void __iomem *base;
> +	phys_addr_t phys_addr;
>  	int cont_id;
>  	int irq;
>  	bool irq_disabled;
> @@ -214,8 +239,18 @@ struct tegra_i2c_dev {
>  	u16 clk_divisor_non_hs_mode;
>  	bool is_multimaster_mode;
>  	spinlock_t xfer_lock;
> +	bool has_dma;
> +	struct dma_chan *tx_dma_chan;
> +	struct dma_chan *rx_dma_chan;
> +	dma_addr_t dma_phys;
> +	u32 *dma_buf;
> +	unsigned int dma_buf_size;
> +	bool is_curr_dma_xfer;
> +	struct completion dma_complete;
>  };
>  
> +static struct dma_chan *chan;
> +
>  static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val,
>  		       unsigned long reg)
>  {
> @@ -282,6 +317,75 @@ static void tegra_i2c_unmask_irq(struct tegra_i2c_dev *i2c_dev, u32 mask)
>  	i2c_writel(i2c_dev, int_mask, I2C_INT_MASK);
>  }
>  
> +static void tegra_i2c_dma_complete(void *args)
> +{
> +	struct tegra_i2c_dev *i2c_dev = args;
> +
> +	complete(&i2c_dev->dma_complete);
> +}
> +
> +static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
> +{
> +	struct dma_async_tx_descriptor *dma_desc;
> +	enum dma_transfer_direction dir;
> +
> +	dev_dbg(i2c_dev->dev, "Starting DMA for length: %zu\n", len);
> +	reinit_completion(&i2c_dev->dma_complete);
> +	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
> +	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
> +					       len, dir, DMA_PREP_INTERRUPT |
> +					       DMA_CTRL_ACK);
> +	if (!dma_desc) {
> +		dev_err(i2c_dev->dev, "Failed to get DMA descriptor\n");
> +		return -EIO;
> +	}
> +
> +	dma_desc->callback = tegra_i2c_dma_complete;
> +	dma_desc->callback_param = i2c_dev;
> +	dmaengine_submit(dma_desc);
> +	dma_async_issue_pending(chan);
> +	return 0;
> +}
> +
> +static int tegra_i2c_init_dma_param(struct tegra_i2c_dev *i2c_dev,
> +				    bool dma_to_memory)
> +{
> +	struct dma_chan *dma_chan;
> +	u32 *dma_buf;
> +	dma_addr_t dma_phys;
> +	int ret;
> +	const char *chan_name = dma_to_memory ? "rx" : "tx";
> +
> +	dma_chan = dma_request_slave_channel_reason(i2c_dev->dev, chan_name);
> +	if (IS_ERR(dma_chan))
> +		return PTR_ERR(dma_chan);
> +
> +	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +				     &dma_phys, GFP_KERNEL);
> +
> +	if (!dma_buf) {
> +		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
> +		ret = -ENOMEM;
> +		goto scrub;
> +	}
> +
> +	if (dma_to_memory)
> +		i2c_dev->rx_dma_chan = dma_chan;
> +	else
> +		i2c_dev->tx_dma_chan = dma_chan;
> +
> +	i2c_dev->dma_buf = dma_buf;
> +	i2c_dev->dma_phys = dma_phys;
> +
> +	return 0;
> +
> +scrub:
> +	dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +			  dma_buf, dma_phys);

Will dma_free_coherent() handle NULL ptr which will happen when dma_buf allocation fails for the RX init'ing?

> +	dma_release_channel(dma_chan);
> +	return ret;
> +}
> +
>  static int tegra_i2c_flush_fifos(struct tegra_i2c_dev *i2c_dev)
>  {
>  	unsigned long timeout = jiffies + HZ;
> @@ -641,25 +745,45 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  		goto err;
>  	}
>  
> -	if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_empty_rx_fifo(i2c_dev);
> -		else
> -			BUG();
> -	}
> +	if (!i2c_dev->is_curr_dma_xfer) {
> +		if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_empty_rx_fifo(i2c_dev);
> +			else
> +				BUG();
> +		}
>  
> -	if (!i2c_dev->msg_read && (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_fill_tx_fifo(i2c_dev);
> -		else
> -			tegra_i2c_mask_irq(i2c_dev, I2C_INT_TX_FIFO_DATA_REQ);
> +		if (!i2c_dev->msg_read &&
> +		   (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_fill_tx_fifo(i2c_dev);
> +			else
> +				tegra_i2c_mask_irq(i2c_dev,
> +						   I2C_INT_TX_FIFO_DATA_REQ);
> +		}
>  	}
>  
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> -	if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +	if (status & I2C_INT_ALL_PACKETS_XFER_COMPLETE) {
> +	/*
> +	 * During message read XFER_COMPLETE interrupt is triggered prior to
> +	 * DMA complete notification and during message write XFER_COMPLETE
> +	 * interrupt is triggered after DMA complete notification.
> +	 * PACKETS_XFER_COMPLETE indicates completion of all bytes of transfer.
> +	 * so forcing msg_buf_remaining to 0.
> +	 */
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
> +		status |= I2C_INT_PACKET_XFER_COMPLETE;
> +		i2c_writel(i2c_dev, status, I2C_INT_STATUS);
> +		if (!i2c_dev->msg_buf_remaining)
> +			complete(&i2c_dev->msg_complete);
> +	} else if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
>  		BUG_ON(i2c_dev->msg_buf_remaining);
>  		complete(&i2c_dev->msg_complete);
>  	}
> @@ -668,17 +792,68 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  	/* An error occurred, mask all interrupts */
>  	tegra_i2c_mask_irq(i2c_dev, I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST |
>  		I2C_INT_PACKET_XFER_COMPLETE | I2C_INT_TX_FIFO_DATA_REQ |
> -		I2C_INT_RX_FIFO_DATA_REQ);
> +		I2C_INT_RX_FIFO_DATA_REQ | I2C_INT_ALL_PACKETS_XFER_COMPLETE);
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> +	if (i2c_dev->is_curr_dma_xfer) {
> +		dmaengine_terminate_all(chan);
> +		complete(&i2c_dev->dma_complete);
> +	}
> +
>  	complete(&i2c_dev->msg_complete);
>  done:
>  	spin_unlock(&i2c_dev->xfer_lock);
>  	return IRQ_HANDLED;
>  }
>  
> +static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
> +				       size_t len, int direction)
> +{
> +	u32 val, reg;
> +	u8 dma_burst = 0;
> +	struct dma_slave_config dma_sconfig;
> +
> +	if (i2c_dev->hw->has_mst_fifo)
> +		reg = I2C_MST_FIFO_CONTROL;
> +	else
> +		reg = I2C_FIFO_CONTROL;
> +	val = i2c_readl(i2c_dev, reg);
> +
> +	if (len & 0xF)
> +		dma_burst = 1;
> +	else if (len & 0x10)
> +		dma_burst = 4;
> +	else
> +		dma_burst = 8;

I'm still wondering whether dma_burst < 8 negatively affects transfer efficiency. Could you please clarify?

> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_TX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_TX_TRIG(dma_burst);
> +	} else {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_RX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
> +	}
> +	i2c_writel(i2c_dev, val, reg);
> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		dma_sconfig.dst_addr = i2c_dev->phys_addr + I2C_TX_FIFO;
> +		dma_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.dst_maxburst = dma_burst;
> +	} else {
> +		dma_sconfig.src_addr = i2c_dev->phys_addr + I2C_RX_FIFO;
> +		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.src_maxburst = dma_burst;
> +	}
> +
> +	dmaengine_slave_config(chan, &dma_sconfig);
> +}
> +
>  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	struct i2c_msg *msg, enum msg_end_type end_state)
>  {
> @@ -688,6 +863,9 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	unsigned long flags;
>  	u16 xfer_time = 100;
>  	size_t xfer_size = 0;
> +	u32 *buffer = 0;
> +	int ret = 0;
> +	bool dma = false;
>  
>  	if (msg->flags & I2C_M_RD)
>  		xfer_size = ALIGN(msg->len, BYTES_PER_FIFO_WORD);
> @@ -698,6 +876,11 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	xfer_time += DIV_ROUND_CLOSEST((xfer_size * 9) * 1000,
>  					i2c_dev->bus_clk_rate);
>  
> +	dma = ((xfer_size > I2C_PIO_MODE_MAX_LEN) &&
> +		i2c_dev->tx_dma_chan && i2c_dev->rx_dma_chan);

There is no need to allocate resources for DMA until dma==true, let's postpone channels requesting and dma_buf allocation until they are really needed.

Please factor out DMA resorces allocation into something like this:

int tegra_i2c_init_dma_resorces(struct tegra_i2c_dev *i2c_dev)
{
	struct dma_chan *dma_rx, *dma_tx;
	dma_addr_t dma_phys;
	u32 *dma_buf;
	int err;

	if (i2c_dev->dma_buf)
		return 0;

	if (!i2c_dev->has_dma)
		return -EINVAL;

	dma_rx = dma_request_slave_channel_reason(i2c_dev->dev, "rx");
	if (IS_ERR(dma_rx))
		return PTR_ERR(dma_rx);

	dma_tx = dma_request_slave_channel_reason(i2c_dev->dev, "tx");
	if (IS_ERR(dma_tx)) {
		err = PTR_ERR(dma_tx);
		goto err_release_rx;
	}

	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
				     &dma_phys, GFP_KERNEL | __GFP_NOWARN);
	if (!dma_buf) {
		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
		err = -ENOMEM;
		goto err_release_tx;
	}

	i2c_dev->dma_buf = dma_buf;
	i2c_dev->dma_phys = dma_phys;
	i2c_dev->rx_dma_chan = dma_rx;
	i2c_dev->tx_dma_chan = dma_tx;

err_release_tx:
	dma_release_channel(dma_tx);
err_release_rx:
	dma_release_channel(dma_rx);

	return err;
}

...

and then:

	dma = (xfer_size > I2C_PIO_MODE_MAX_LEN);
	if (dma) {
		err = tegra_i2c_init_dma_resorces(i2c_dev);
		if (err)
			dma = false;
	}

> +
> +	i2c_dev->is_curr_dma_xfer = dma;
> +
>  	tegra_i2c_flush_fifos(i2c_dev);
>  
>  	i2c_dev->msg_buf = msg->buf;
> @@ -711,14 +894,50 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	int_mask = I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST;
>  	tegra_i2c_unmask_irq(i2c_dev, int_mask);
>  
> +	if (dma) {
> +		if (i2c_dev->msg_read) {
> +			chan = i2c_dev->rx_dma_chan;
> +			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
> +						   DATA_DMA_DIR_RX);
> +			/* make the dma buffer to read by dma */
> +			dma_sync_single_for_device(i2c_dev->dev,
> +						   i2c_dev->dma_phys,
> +						   i2c_dev->dma_buf_size,

Why not use the size of transfer for the syncing?

> +						   DMA_FROM_DEVICE);
> +			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
> +			if (ret < 0) {
> +				dev_err(i2c_dev->dev,
> +					"Starting RX DMA failed, err %d\n",
> +					ret);
> +				goto exit;
> +			}
> +		} else {
> +			chan = i2c_dev->tx_dma_chan;
> +			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
> +						   DATA_DMA_DIR_TX);
> +			/* Make the dma buffer to read by cpu */
> +			dma_sync_single_for_cpu(i2c_dev->dev,
> +						i2c_dev->dma_phys,
> +						i2c_dev->dma_buf_size,> +						DMA_TO_DEVICE);
> +			buffer = (u32 *)i2c_dev->dma_buf;
> +		}
> +	}
> +
>  	packet_header = (0 << PACKET_HEADER0_HEADER_SIZE_SHIFT) |
>  			PACKET_HEADER0_PROTOCOL_I2C |
>  			(i2c_dev->cont_id << PACKET_HEADER0_CONT_ID_SHIFT) |
>  			(1 << PACKET_HEADER0_PACKET_ID_SHIFT);
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
>  
>  	packet_header = msg->len - 1;
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
>  
>  	packet_header = I2C_HEADER_IE_ENABLE;
>  	if (end_state == MSG_END_CONTINUE)
> @@ -735,29 +954,85 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  		packet_header |= I2C_HEADER_CONT_ON_NAK;
>  	if (msg->flags & I2C_M_RD)
>  		packet_header |= I2C_HEADER_READ;
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> -
> -	if (!(msg->flags & I2C_M_RD))
> -		tegra_i2c_fill_tx_fifo(i2c_dev);
> -
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +
> +	if (!i2c_dev->msg_read) {
> +		if (dma) {
> +			memcpy(buffer, msg->buf, msg->len);
> +			/* make the dma buffer to read by dma */
> +			dma_sync_single_for_device(i2c_dev->dev,
> +						   i2c_dev->dma_phys,
> +						   i2c_dev->dma_buf_size,
> +						   DMA_TO_DEVICE);
> +			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
> +			if (ret < 0) {
> +				dev_err(i2c_dev->dev,
> +					"Starting TX DMA failed, err %d\n",
> +					ret);
> +				goto exit;
> +			}
> +		} else {
> +			tegra_i2c_fill_tx_fifo(i2c_dev);
> +		}
> +	}
>  	if (i2c_dev->hw->has_per_pkt_xfer_complete_irq)
>  		int_mask |= I2C_INT_PACKET_XFER_COMPLETE;
> -	if (msg->flags & I2C_M_RD)
> -		int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
> -	else if (i2c_dev->msg_buf_remaining)
> -		int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
> +	if (dma) {
> +		int_mask |= I2C_INT_ALL_PACKETS_XFER_COMPLETE;
> +	} else {
> +		if (msg->flags & I2C_M_RD)
> +			int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
> +		else if (i2c_dev->msg_buf_remaining)
> +			int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
> +	}
>  
>  	tegra_i2c_unmask_irq(i2c_dev, int_mask);
> -	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
>  	dev_dbg(i2c_dev->dev, "unmasked irq: %02x\n",
>  		i2c_readl(i2c_dev, I2C_INT_MASK));
>  
> +exit:
> +	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
> +	if (ret)
> +		return ret;
> +
> +	if (dma) {
> +		time_left = wait_for_completion_timeout(
> +						&i2c_dev->dma_complete,
> +						TEGRA_I2C_TIMEOUT(xfer_time));
> +
> +		if (time_left == 0) {
> +			dev_err(i2c_dev->dev, "DMA transfer timeout\n");
> +			dmaengine_terminate_all(chan);
> +			tegra_i2c_init(i2c_dev);
> +			return -ETIMEDOUT;
> +		}
> +
> +		if (i2c_dev->msg_read) {
> +			if (likely(i2c_dev->msg_err == I2C_ERR_NONE)) {
> +				dma_sync_single_for_cpu(i2c_dev->dev,
> +							i2c_dev->dma_phys,
> +							i2c_dev->dma_buf_size,
> +							DMA_FROM_DEVICE);
> +
> +				memcpy(i2c_dev->msg_buf, i2c_dev->dma_buf,
> +					msg->len);
> +			}
> +		}
> +	}
> +
>  	time_left = wait_for_completion_timeout(&i2c_dev->msg_complete,
>  						TEGRA_I2C_TIMEOUT(xfer_time));
>  	tegra_i2c_mask_irq(i2c_dev, int_mask);
>  
>  	if (time_left == 0) {
>  		dev_err(i2c_dev->dev, "i2c transfer timed out\n");
> +		if (dma) {
> +			dmaengine_terminate_all(chan);
> +			complete(&i2c_dev->dma_complete);
> +		}
>  
>  		tegra_i2c_init(i2c_dev);
>  		return -ETIMEDOUT;
> @@ -835,6 +1110,8 @@ static void tegra_i2c_parse_dt(struct tegra_i2c_dev *i2c_dev)
>  
>  	i2c_dev->is_multimaster_mode = of_property_read_bool(np,
>  			"multi-master");
> +
> +	i2c_dev->has_dma = of_property_read_bool(np, "dmas");
>  }
>  
>  static const struct i2c_algorithm tegra_i2c_algo = {
> @@ -947,11 +1224,13 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  	struct clk *div_clk;
>  	struct clk *fast_clk;
>  	void __iomem *base;
> +	phys_addr_t phys_addr;
>  	int irq;
>  	int ret = 0;
>  	int clk_multiplier = I2C_CLK_MULTIPLIER_STD_FAST_MODE;
>  
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	phys_addr = res->start;
>  	base = devm_ioremap_resource(&pdev->dev, res);
>  	if (IS_ERR(base))
>  		return PTR_ERR(base);
> @@ -974,12 +1253,14 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  		return -ENOMEM;
>  
>  	i2c_dev->base = base;
> +	i2c_dev->phys_addr = phys_addr;
>  	i2c_dev->div_clk = div_clk;
>  	i2c_dev->adapter.algo = &tegra_i2c_algo;
>  	i2c_dev->adapter.quirks = &tegra_i2c_quirks;
>  	i2c_dev->irq = irq;
>  	i2c_dev->cont_id = pdev->id;
>  	i2c_dev->dev = &pdev->dev;
> +	i2c_dev->dma_buf_size = i2c_dev->adapter.quirks->max_write_len;
>  
>  	i2c_dev->rst = devm_reset_control_get_exclusive(&pdev->dev, "i2c");
>  	if (IS_ERR(i2c_dev->rst)) {
> @@ -994,6 +1275,7 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  						  "nvidia,tegra20-i2c-dvc");
>  	init_completion(&i2c_dev->msg_complete);
>  	spin_lock_init(&i2c_dev->xfer_lock);
> +	init_completion(&i2c_dev->dma_complete);
>  
>  	if (!i2c_dev->hw->has_single_clk_source) {
>  		fast_clk = devm_clk_get(&pdev->dev, "fast-clk");
> @@ -1053,6 +1335,15 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  		}
>  	}
>  
> +	if (i2c_dev->has_dma) {
> +		ret = tegra_i2c_init_dma_param(i2c_dev, true);
> +		if (ret == -EPROBE_DEFER)
> +			goto disable_div_clk;

So now dma_buf has been allocated by initing the RX channel.

> +		ret = tegra_i2c_init_dma_param(i2c_dev, false);

And now dma_buf will be allocated second time for the TX?

> +		if (ret == -EPROBE_DEFER)
> +			goto disable_div_clk;

Where is RX channel releasing when TX init fails?

> +	}

No channels / dma_buf releasing in tegra_i2c_remove()? 

> +
>  	ret = tegra_i2c_init(i2c_dev);
>  	if (ret) {
>  		dev_err(&pdev->dev, "Failed to initialize i2c controller\n");
> 

Please collect performance stats before posting new version of the patch.
Dmitry Osipenko Jan. 26, 2019, 5:29 p.m. UTC | #2
26.01.2019 6:57, Sowjanya Komatineni пишет:
> This patch adds DMA support for Tegra I2C.
> 
> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
> transfer size of the max FIFO depth and DMA mode is used for
> transfer size higher than max FIFO depth to save CPU overhead.
> 
> PIO mode needs full intervention of CPU to fill or empty FIFO's
> and also need to service multiple data requests interrupt for the
> same transaction adding overhead on CPU for large transfers.
> 
> DMA mode is helpful for Large transfers during downloading or
> uploading FW over I2C to some external devices.
> 
> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> 
> ---
>  [V3] : Updated without additional buffer allocation.
>  [V2] : Updated based on V1 review feedback along with code cleanup for
> 	proper implementation of DMA.
> 
>  drivers/i2c/busses/i2c-tegra.c | 341 ++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 316 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
> index 3dcbc9960d9d..452358a77400 100644
> --- a/drivers/i2c/busses/i2c-tegra.c
> +++ b/drivers/i2c/busses/i2c-tegra.c
> @@ -8,6 +8,9 @@
>  
>  #include <linux/clk.h>
>  #include <linux/delay.h>
> +#include <linux/dmaengine.h>
> +#include <linux/dmapool.h>
> +#include <linux/dma-mapping.h>
>  #include <linux/err.h>
>  #include <linux/i2c.h>
>  #include <linux/init.h>
> @@ -45,6 +48,8 @@
>  #define I2C_FIFO_CONTROL_RX_FLUSH		BIT(0)
>  #define I2C_FIFO_CONTROL_TX_TRIG_SHIFT		5
>  #define I2C_FIFO_CONTROL_RX_TRIG_SHIFT		2
> +#define I2C_FIFO_CONTROL_TX_TRIG(x)		(((x) - 1) << 5)
> +#define I2C_FIFO_CONTROL_RX_TRIG(x)		(((x) - 1) << 2)
>  #define I2C_FIFO_STATUS				0x060
>  #define I2C_FIFO_STATUS_TX_MASK			0xF0
>  #define I2C_FIFO_STATUS_TX_SHIFT		4
> @@ -119,6 +124,16 @@
>  /* Packet header size in bytes */
>  #define I2C_PACKET_HEADER_SIZE			12
>  
> +#define DATA_DMA_DIR_TX				(1 << 0)
> +#define DATA_DMA_DIR_RX				(1 << 1)
> +
> +/*
> + * Upto I2C_PIO_MODE_MAX_LEN bytes, controller will use PIO mode,
> + * above this, controller will use DMA to fill FIFO.
> + * MAX PIO len is 20 bytes excluding packet header.
> + */
> +#define I2C_PIO_MODE_MAX_LEN			20
> +
>  /*
>   * msg_end_type: The bus control which need to be send at end of transfer.
>   * @MSG_END_STOP: Send stop pulse at end of transfer.
> @@ -179,6 +194,7 @@ struct tegra_i2c_hw_feature {
>   * @fast_clk: clock reference for fast clock of I2C controller
>   * @rst: reset control for the I2C controller
>   * @base: ioremapped registers cookie
> + * @phys_addr: Physical address of I2C base address to use for DMA configuration
>   * @cont_id: I2C controller ID, used for packet header
>   * @irq: IRQ number of transfer complete interrupt
>   * @irq_disabled: used to track whether or not the interrupt is enabled
> @@ -192,6 +208,14 @@ struct tegra_i2c_hw_feature {
>   * @clk_divisor_non_hs_mode: clock divider for non-high-speed modes
>   * @is_multimaster_mode: track if I2C controller is in multi-master mode
>   * @xfer_lock: lock to serialize transfer submission and processing
> + * @has_dma: indicated if controller supports DMA
> + * @tx_dma_chan: DMA transmit channel
> + * @rx_dma_chan: DMA receive channel
> + * @dma_phys: handle to DMA resources
> + * @dma_buf: pointer to allocated DMA buffer
> + * @dma_buf_size: DMA buffer size
> + * @is_curr_dma_xfer: indicates active DMA transfer
> + * @dma_complete: DMA completion notifier
>   */
>  struct tegra_i2c_dev {
>  	struct device *dev;
> @@ -201,6 +225,7 @@ struct tegra_i2c_dev {
>  	struct clk *fast_clk;
>  	struct reset_control *rst;
>  	void __iomem *base;
> +	phys_addr_t phys_addr;
>  	int cont_id;
>  	int irq;
>  	bool irq_disabled;
> @@ -214,8 +239,18 @@ struct tegra_i2c_dev {
>  	u16 clk_divisor_non_hs_mode;
>  	bool is_multimaster_mode;
>  	spinlock_t xfer_lock;
> +	bool has_dma;
> +	struct dma_chan *tx_dma_chan;
> +	struct dma_chan *rx_dma_chan;
> +	dma_addr_t dma_phys;
> +	u32 *dma_buf;
> +	unsigned int dma_buf_size;
> +	bool is_curr_dma_xfer;
> +	struct completion dma_complete;
>  };
>  
> +static struct dma_chan *chan;
> +
>  static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val,
>  		       unsigned long reg)
>  {
> @@ -282,6 +317,75 @@ static void tegra_i2c_unmask_irq(struct tegra_i2c_dev *i2c_dev, u32 mask)
>  	i2c_writel(i2c_dev, int_mask, I2C_INT_MASK);
>  }
>  
> +static void tegra_i2c_dma_complete(void *args)
> +{
> +	struct tegra_i2c_dev *i2c_dev = args;
> +
> +	complete(&i2c_dev->dma_complete);
> +}
> +
> +static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
> +{
> +	struct dma_async_tx_descriptor *dma_desc;
> +	enum dma_transfer_direction dir;
> +
> +	dev_dbg(i2c_dev->dev, "Starting DMA for length: %zu\n", len);
> +	reinit_completion(&i2c_dev->dma_complete);
> +	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
> +	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
> +					       len, dir, DMA_PREP_INTERRUPT |
> +					       DMA_CTRL_ACK);
> +	if (!dma_desc) {
> +		dev_err(i2c_dev->dev, "Failed to get DMA descriptor\n");
> +		return -EIO;
> +	}
> +
> +	dma_desc->callback = tegra_i2c_dma_complete;
> +	dma_desc->callback_param = i2c_dev;
> +	dmaengine_submit(dma_desc);
> +	dma_async_issue_pending(chan);
> +	return 0;
> +}
> +
> +static int tegra_i2c_init_dma_param(struct tegra_i2c_dev *i2c_dev,
> +				    bool dma_to_memory)
> +{
> +	struct dma_chan *dma_chan;
> +	u32 *dma_buf;
> +	dma_addr_t dma_phys;
> +	int ret;
> +	const char *chan_name = dma_to_memory ? "rx" : "tx";
> +
> +	dma_chan = dma_request_slave_channel_reason(i2c_dev->dev, chan_name);
> +	if (IS_ERR(dma_chan))
> +		return PTR_ERR(dma_chan);
> +
> +	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +				     &dma_phys, GFP_KERNEL);
> +
> +	if (!dma_buf) {
> +		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
> +		ret = -ENOMEM;
> +		goto scrub;
> +	}
> +
> +	if (dma_to_memory)
> +		i2c_dev->rx_dma_chan = dma_chan;
> +	else
> +		i2c_dev->tx_dma_chan = dma_chan;
> +
> +	i2c_dev->dma_buf = dma_buf;
> +	i2c_dev->dma_phys = dma_phys;
> +
> +	return 0;
> +
> +scrub:
> +	dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +			  dma_buf, dma_phys);
> +	dma_release_channel(dma_chan);
> +	return ret;
> +}
> +
>  static int tegra_i2c_flush_fifos(struct tegra_i2c_dev *i2c_dev)
>  {
>  	unsigned long timeout = jiffies + HZ;
> @@ -641,25 +745,45 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  		goto err;
>  	}
>  
> -	if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_empty_rx_fifo(i2c_dev);
> -		else
> -			BUG();
> -	}
> +	if (!i2c_dev->is_curr_dma_xfer) {
> +		if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_empty_rx_fifo(i2c_dev);
> +			else
> +				BUG();
> +		}
>  
> -	if (!i2c_dev->msg_read && (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_fill_tx_fifo(i2c_dev);
> -		else
> -			tegra_i2c_mask_irq(i2c_dev, I2C_INT_TX_FIFO_DATA_REQ);
> +		if (!i2c_dev->msg_read &&
> +		   (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_fill_tx_fifo(i2c_dev);
> +			else
> +				tegra_i2c_mask_irq(i2c_dev,
> +						   I2C_INT_TX_FIFO_DATA_REQ);
> +		}
>  	}
>  
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> -	if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +	if (status & I2C_INT_ALL_PACKETS_XFER_COMPLETE) {
> +	/*
> +	 * During message read XFER_COMPLETE interrupt is triggered prior to
> +	 * DMA complete notification and during message write XFER_COMPLETE
> +	 * interrupt is triggered after DMA complete notification.
> +	 * PACKETS_XFER_COMPLETE indicates completion of all bytes of transfer.
> +	 * so forcing msg_buf_remaining to 0.
> +	 */
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
> +		status |= I2C_INT_PACKET_XFER_COMPLETE;
> +		i2c_writel(i2c_dev, status, I2C_INT_STATUS);
> +		if (!i2c_dev->msg_buf_remaining)
> +			complete(&i2c_dev->msg_complete);
> +	} else if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
>  		BUG_ON(i2c_dev->msg_buf_remaining);
>  		complete(&i2c_dev->msg_complete);
>  	}
> @@ -668,17 +792,68 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  	/* An error occurred, mask all interrupts */
>  	tegra_i2c_mask_irq(i2c_dev, I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST |
>  		I2C_INT_PACKET_XFER_COMPLETE | I2C_INT_TX_FIFO_DATA_REQ |
> -		I2C_INT_RX_FIFO_DATA_REQ);
> +		I2C_INT_RX_FIFO_DATA_REQ | I2C_INT_ALL_PACKETS_XFER_COMPLETE);
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> +	if (i2c_dev->is_curr_dma_xfer) {
> +		dmaengine_terminate_all(chan);
> +		complete(&i2c_dev->dma_complete);
> +	}
> +
>  	complete(&i2c_dev->msg_complete);
>  done:
>  	spin_unlock(&i2c_dev->xfer_lock);
>  	return IRQ_HANDLED;
>  }
>  
> +static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
> +				       size_t len, int direction)
> +{
> +	u32 val, reg;
> +	u8 dma_burst = 0;
> +	struct dma_slave_config dma_sconfig;
> +
> +	if (i2c_dev->hw->has_mst_fifo)
> +		reg = I2C_MST_FIFO_CONTROL;
> +	else
> +		reg = I2C_FIFO_CONTROL;
> +	val = i2c_readl(i2c_dev, reg);
> +
> +	if (len & 0xF)
> +		dma_burst = 1;
> +	else if (len & 0x10)
> +		dma_burst = 4;
> +	else
> +		dma_burst = 8;
> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_TX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_TX_TRIG(dma_burst);
> +	} else {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_RX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
> +	}
> +	i2c_writel(i2c_dev, val, reg);
> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		dma_sconfig.dst_addr = i2c_dev->phys_addr + I2C_TX_FIFO;
> +		dma_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.dst_maxburst = dma_burst;
> +	} else {
> +		dma_sconfig.src_addr = i2c_dev->phys_addr + I2C_RX_FIFO;
> +		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.src_maxburst = dma_burst;
> +	}
> +
> +	dmaengine_slave_config(chan, &dma_sconfig);
> +}
> +
>  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	struct i2c_msg *msg, enum msg_end_type end_state)
>  {
> @@ -688,6 +863,9 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	unsigned long flags;
>  	u16 xfer_time = 100;
>  	size_t xfer_size = 0;
> +	u32 *buffer = 0;
> +	int ret = 0;
> +	bool dma = false;
>  
>  	if (msg->flags & I2C_M_RD)
>  		xfer_size = ALIGN(msg->len, BYTES_PER_FIFO_WORD);
> @@ -698,6 +876,11 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	xfer_time += DIV_ROUND_CLOSEST((xfer_size * 9) * 1000,
>  					i2c_dev->bus_clk_rate);
>  
> +	dma = ((xfer_size > I2C_PIO_MODE_MAX_LEN) &&
> +		i2c_dev->tx_dma_chan && i2c_dev->rx_dma_chan);
> +
> +	i2c_dev->is_curr_dma_xfer = dma;
> +
>  	tegra_i2c_flush_fifos(i2c_dev);
>  
>  	i2c_dev->msg_buf = msg->buf;
> @@ -711,14 +894,50 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	int_mask = I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST;
>  	tegra_i2c_unmask_irq(i2c_dev, int_mask);
>  
> +	if (dma) {
> +		if (i2c_dev->msg_read) {
> +			chan = i2c_dev->rx_dma_chan;
> +			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
> +						   DATA_DMA_DIR_RX);
> +			/* make the dma buffer to read by dma */
> +			dma_sync_single_for_device(i2c_dev->dev,
> +						   i2c_dev->dma_phys,
> +						   i2c_dev->dma_buf_size,
> +						   DMA_FROM_DEVICE);
> +			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
> +			if (ret < 0) {
> +				dev_err(i2c_dev->dev,
> +					"Starting RX DMA failed, err %d\n",
> +					ret);
> +				goto exit;
> +			}
> +		} else {
> +			chan = i2c_dev->tx_dma_chan;
> +			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
> +						   DATA_DMA_DIR_TX);
> +			/* Make the dma buffer to read by cpu */
> +			dma_sync_single_for_cpu(i2c_dev->dev,
> +						i2c_dev->dma_phys,
> +						i2c_dev->dma_buf_size,
> +						DMA_TO_DEVICE);
> +			buffer = (u32 *)i2c_dev->dma_buf;
> +		}
> +	}
> +
>  	packet_header = (0 << PACKET_HEADER0_HEADER_SIZE_SHIFT) |
>  			PACKET_HEADER0_PROTOCOL_I2C |
>  			(i2c_dev->cont_id << PACKET_HEADER0_CONT_ID_SHIFT) |
>  			(1 << PACKET_HEADER0_PACKET_ID_SHIFT);
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
>  
>  	packet_header = msg->len - 1;
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
>  
>  	packet_header = I2C_HEADER_IE_ENABLE;
>  	if (end_state == MSG_END_CONTINUE)
> @@ -735,29 +954,85 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  		packet_header |= I2C_HEADER_CONT_ON_NAK;
>  	if (msg->flags & I2C_M_RD)
>  		packet_header |= I2C_HEADER_READ;
> -	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> -
> -	if (!(msg->flags & I2C_M_RD))
> -		tegra_i2c_fill_tx_fifo(i2c_dev);
> -
> +	if (dma && !i2c_dev->msg_read)
> +		*buffer++ = packet_header;
> +	else
> +		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
> +
> +	if (!i2c_dev->msg_read) {
> +		if (dma) {
> +			memcpy(buffer, msg->buf, msg->len);
> +			/* make the dma buffer to read by dma */
> +			dma_sync_single_for_device(i2c_dev->dev,
> +						   i2c_dev->dma_phys,
> +						   i2c_dev->dma_buf_size,
> +						   DMA_TO_DEVICE);
> +			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
> +			if (ret < 0) {
> +				dev_err(i2c_dev->dev,
> +					"Starting TX DMA failed, err %d\n",
> +					ret);
> +				goto exit;
> +			}
> +		} else {
> +			tegra_i2c_fill_tx_fifo(i2c_dev);
> +		}
> +	}
>  	if (i2c_dev->hw->has_per_pkt_xfer_complete_irq)
>  		int_mask |= I2C_INT_PACKET_XFER_COMPLETE;
> -	if (msg->flags & I2C_M_RD)
> -		int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
> -	else if (i2c_dev->msg_buf_remaining)
> -		int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
> +	if (dma) {
> +		int_mask |= I2C_INT_ALL_PACKETS_XFER_COMPLETE;
> +	} else {
> +		if (msg->flags & I2C_M_RD)
> +			int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
> +		else if (i2c_dev->msg_buf_remaining)
> +			int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
> +	}
>  
>  	tegra_i2c_unmask_irq(i2c_dev, int_mask);
> -	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
>  	dev_dbg(i2c_dev->dev, "unmasked irq: %02x\n",
>  		i2c_readl(i2c_dev, I2C_INT_MASK));
>  
> +exit:
> +	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
> +	if (ret)
> +		return ret;
> +
> +	if (dma) {
> +		time_left = wait_for_completion_timeout(
> +						&i2c_dev->dma_complete,
> +						TEGRA_I2C_TIMEOUT(xfer_time));
> +
> +		if (time_left == 0) {
> +			dev_err(i2c_dev->dev, "DMA transfer timeout\n");
> +			dmaengine_terminate_all(chan);
> +			tegra_i2c_init(i2c_dev);
> +			return -ETIMEDOUT;
> +		}
> +
> +		if (i2c_dev->msg_read) {
> +			if (likely(i2c_dev->msg_err == I2C_ERR_NONE)) {
> +				dma_sync_single_for_cpu(i2c_dev->dev,
> +							i2c_dev->dma_phys,
> +							i2c_dev->dma_buf_size,
> +							DMA_FROM_DEVICE);
> +
> +				memcpy(i2c_dev->msg_buf, i2c_dev->dma_buf,
> +					msg->len);
> +			}
> +		}
> +	}
> +
>  	time_left = wait_for_completion_timeout(&i2c_dev->msg_complete,
>  						TEGRA_I2C_TIMEOUT(xfer_time));
>  	tegra_i2c_mask_irq(i2c_dev, int_mask);
>  
>  	if (time_left == 0) {
>  		dev_err(i2c_dev->dev, "i2c transfer timed out\n");
> +		if (dma) {
> +			dmaengine_terminate_all(chan);
> +			complete(&i2c_dev->dma_complete);
> +		}
>  
>  		tegra_i2c_init(i2c_dev);
>  		return -ETIMEDOUT;
> @@ -835,6 +1110,8 @@ static void tegra_i2c_parse_dt(struct tegra_i2c_dev *i2c_dev)
>  
>  	i2c_dev->is_multimaster_mode = of_property_read_bool(np,
>  			"multi-master");
> +
> +	i2c_dev->has_dma = of_property_read_bool(np, "dmas");
>  }
>  
>  static const struct i2c_algorithm tegra_i2c_algo = {
> @@ -947,11 +1224,13 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  	struct clk *div_clk;
>  	struct clk *fast_clk;
>  	void __iomem *base;
> +	phys_addr_t phys_addr;
>  	int irq;
>  	int ret = 0;
>  	int clk_multiplier = I2C_CLK_MULTIPLIER_STD_FAST_MODE;
>  
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	phys_addr = res->start;
>  	base = devm_ioremap_resource(&pdev->dev, res);
>  	if (IS_ERR(base))
>  		return PTR_ERR(base);
> @@ -974,12 +1253,14 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  		return -ENOMEM;
>  
>  	i2c_dev->base = base;
> +	i2c_dev->phys_addr = phys_addr;
>  	i2c_dev->div_clk = div_clk;
>  	i2c_dev->adapter.algo = &tegra_i2c_algo;
>  	i2c_dev->adapter.quirks = &tegra_i2c_quirks;
>  	i2c_dev->irq = irq;
>  	i2c_dev->cont_id = pdev->id;
>  	i2c_dev->dev = &pdev->dev;
> +	i2c_dev->dma_buf_size = i2c_dev->adapter.quirks->max_write_len;
>  
>  	i2c_dev->rst = devm_reset_control_get_exclusive(&pdev->dev, "i2c");
>  	if (IS_ERR(i2c_dev->rst)) {
> @@ -994,6 +1275,7 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  						  "nvidia,tegra20-i2c-dvc");
>  	init_completion(&i2c_dev->msg_complete);
>  	spin_lock_init(&i2c_dev->xfer_lock);
> +	init_completion(&i2c_dev->dma_complete);
>  
>  	if (!i2c_dev->hw->has_single_clk_source) {
>  		fast_clk = devm_clk_get(&pdev->dev, "fast-clk");
> @@ -1053,6 +1335,15 @@ static int tegra_i2c_probe(struct platform_device *pdev)
>  		}
>  	}
>  
> +	if (i2c_dev->has_dma) {
> +		ret = tegra_i2c_init_dma_param(i2c_dev, true);
> +		if (ret == -EPROBE_DEFER)
> +			goto disable_div_clk;
> +		ret = tegra_i2c_init_dma_param(i2c_dev, false);
> +		if (ret == -EPROBE_DEFER)
> +			goto disable_div_clk;
> +	}
> +
>  	ret = tegra_i2c_init(i2c_dev);
>  	if (ret) {
>  		dev_err(&pdev->dev, "Failed to initialize i2c controller\n");
> 

This patch doesn't apply, same as for the v2.. any clue why this happens?

# git apply --verbose V3-3-3-i2c-tegra-Add-DMA-Support.patch 
Checking patch drivers/i2c/busses/i2c-tegra.c...
Hunk #4 succeeded at 197 (offset 3 lines).
Hunk #5 succeeded at 211 (offset 3 lines).
Hunk #6 succeeded at 228 (offset 3 lines).
Hunk #7 succeeded at 242 (offset 3 lines).
Hunk #8 succeeded at 320 (offset 3 lines).
Hunk #9 succeeded at 748 (offset 3 lines).
Hunk #10 succeeded at 795 (offset 3 lines).
Hunk #11 succeeded at 866 (offset 3 lines).
Hunk #12 succeeded at 879 (offset 3 lines).
Hunk #13 succeeded at 897 (offset 3 lines).
Hunk #14 succeeded at 957 (offset 3 lines).
Hunk #15 succeeded at 1113 (offset 3 lines).
Hunk #16 succeeded at 1237 (offset 13 lines).
error: while searching for:
                return -ENOMEM;

        i2c_dev->base = base;
        i2c_dev->div_clk = div_clk;
        i2c_dev->adapter.algo = &tegra_i2c_algo;
        i2c_dev->adapter.quirks = &tegra_i2c_quirks;
        i2c_dev->irq = irq;
        i2c_dev->cont_id = pdev->id;
        i2c_dev->dev = &pdev->dev;

        i2c_dev->rst = devm_reset_control_get_exclusive(&pdev->dev, "i2c");
        if (IS_ERR(i2c_dev->rst)) {

error: patch failed: drivers/i2c/busses/i2c-tegra.c:974
error: drivers/i2c/busses/i2c-tegra.c: patch does not apply
Dmitry Osipenko Jan. 26, 2019, 6:16 p.m. UTC | #3
26.01.2019 20:11, Dmitry Osipenko пишет:
> 26.01.2019 6:57, Sowjanya Komatineni пишет:
>> This patch adds DMA support for Tegra I2C.
>>
>> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
>> transfer size of the max FIFO depth and DMA mode is used for
>> transfer size higher than max FIFO depth to save CPU overhead.
>>
>> PIO mode needs full intervention of CPU to fill or empty FIFO's
>> and also need to service multiple data requests interrupt for the
>> same transaction adding overhead on CPU for large transfers.
>>
>> DMA mode is helpful for Large transfers during downloading or
>> uploading FW over I2C to some external devices.
>>
>> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
>>
>> ---
>>  [V3] : Updated without additional buffer allocation.
>>  [V2] : Updated based on V1 review feedback along with code cleanup for
>> 	proper implementation of DMA.
>>
>>  drivers/i2c/busses/i2c-tegra.c | 341 ++++++++++++++++++++++++++++++++++++++---
>>  1 file changed, 316 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
>> index 3dcbc9960d9d..452358a77400 100644
>> --- a/drivers/i2c/busses/i2c-tegra.c
>> +++ b/drivers/i2c/busses/i2c-tegra.c
>> @@ -8,6 +8,9 @@
>>  
>>  #include <linux/clk.h>
>>  #include <linux/delay.h>
>> +#include <linux/dmaengine.h>
>> +#include <linux/dmapool.h>
>> +#include <linux/dma-mapping.h>
>>  #include <linux/err.h>
>>  #include <linux/i2c.h>
>>  #include <linux/init.h>
>> @@ -45,6 +48,8 @@
>>  #define I2C_FIFO_CONTROL_RX_FLUSH		BIT(0)
>>  #define I2C_FIFO_CONTROL_TX_TRIG_SHIFT		5
>>  #define I2C_FIFO_CONTROL_RX_TRIG_SHIFT		2
>> +#define I2C_FIFO_CONTROL_TX_TRIG(x)		(((x) - 1) << 5)
>> +#define I2C_FIFO_CONTROL_RX_TRIG(x)		(((x) - 1) << 2)
>>  #define I2C_FIFO_STATUS				0x060
>>  #define I2C_FIFO_STATUS_TX_MASK			0xF0
>>  #define I2C_FIFO_STATUS_TX_SHIFT		4
>> @@ -119,6 +124,16 @@
>>  /* Packet header size in bytes */
>>  #define I2C_PACKET_HEADER_SIZE			12
>>  
>> +#define DATA_DMA_DIR_TX				(1 << 0)
>> +#define DATA_DMA_DIR_RX				(1 << 1)
>> +
>> +/*
>> + * Upto I2C_PIO_MODE_MAX_LEN bytes, controller will use PIO mode,
>> + * above this, controller will use DMA to fill FIFO.
>> + * MAX PIO len is 20 bytes excluding packet header.
>> + */
>> +#define I2C_PIO_MODE_MAX_LEN			20
>> +
>>  /*
>>   * msg_end_type: The bus control which need to be send at end of transfer.
>>   * @MSG_END_STOP: Send stop pulse at end of transfer.
>> @@ -179,6 +194,7 @@ struct tegra_i2c_hw_feature {
>>   * @fast_clk: clock reference for fast clock of I2C controller
>>   * @rst: reset control for the I2C controller
>>   * @base: ioremapped registers cookie
>> + * @phys_addr: Physical address of I2C base address to use for DMA configuration
>>   * @cont_id: I2C controller ID, used for packet header
>>   * @irq: IRQ number of transfer complete interrupt
>>   * @irq_disabled: used to track whether or not the interrupt is enabled
>> @@ -192,6 +208,14 @@ struct tegra_i2c_hw_feature {
>>   * @clk_divisor_non_hs_mode: clock divider for non-high-speed modes
>>   * @is_multimaster_mode: track if I2C controller is in multi-master mode
>>   * @xfer_lock: lock to serialize transfer submission and processing
>> + * @has_dma: indicated if controller supports DMA
> 
> AFAIK, all Terga's support DMA. Let's change to "@has_dma: can utilize DMA" for clarity.
> 
>> + * @tx_dma_chan: DMA transmit channel
>> + * @rx_dma_chan: DMA receive channel
>> + * @dma_phys: handle to DMA resources
>> + * @dma_buf: pointer to allocated DMA buffer
>> + * @dma_buf_size: DMA buffer size
>> + * @is_curr_dma_xfer: indicates active DMA transfer
>> + * @dma_complete: DMA completion notifier
>>   */
>>  struct tegra_i2c_dev {
>>  	struct device *dev;
>> @@ -201,6 +225,7 @@ struct tegra_i2c_dev {
>>  	struct clk *fast_clk;
>>  	struct reset_control *rst;
>>  	void __iomem *base;
>> +	phys_addr_t phys_addr;
>>  	int cont_id;
>>  	int irq;
>>  	bool irq_disabled;
>> @@ -214,8 +239,18 @@ struct tegra_i2c_dev {
>>  	u16 clk_divisor_non_hs_mode;
>>  	bool is_multimaster_mode;
>>  	spinlock_t xfer_lock;
>> +	bool has_dma;
>> +	struct dma_chan *tx_dma_chan;
>> +	struct dma_chan *rx_dma_chan;
>> +	dma_addr_t dma_phys;
>> +	u32 *dma_buf;
>> +	unsigned int dma_buf_size;
>> +	bool is_curr_dma_xfer;
>> +	struct completion dma_complete;
>>  };
>>  
>> +static struct dma_chan *chan;
>> +
>>  static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val,
>>  		       unsigned long reg)
>>  {
>> @@ -282,6 +317,75 @@ static void tegra_i2c_unmask_irq(struct tegra_i2c_dev *i2c_dev, u32 mask)
>>  	i2c_writel(i2c_dev, int_mask, I2C_INT_MASK);
>>  }
>>  
>> +static void tegra_i2c_dma_complete(void *args)
>> +{
>> +	struct tegra_i2c_dev *i2c_dev = args;
>> +
>> +	complete(&i2c_dev->dma_complete);
>> +}
>> +
>> +static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
>> +{
>> +	struct dma_async_tx_descriptor *dma_desc;
>> +	enum dma_transfer_direction dir;
>> +
>> +	dev_dbg(i2c_dev->dev, "Starting DMA for length: %zu\n", len);
>> +	reinit_completion(&i2c_dev->dma_complete);
>> +	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
>> +	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
>> +					       len, dir, DMA_PREP_INTERRUPT |
>> +					       DMA_CTRL_ACK);
>> +	if (!dma_desc) {
>> +		dev_err(i2c_dev->dev, "Failed to get DMA descriptor\n");
>> +		return -EIO;
>> +	}
>> +
>> +	dma_desc->callback = tegra_i2c_dma_complete;
>> +	dma_desc->callback_param = i2c_dev;
>> +	dmaengine_submit(dma_desc);
>> +	dma_async_issue_pending(chan);
>> +	return 0;
>> +}
>> +
>> +static int tegra_i2c_init_dma_param(struct tegra_i2c_dev *i2c_dev,
>> +				    bool dma_to_memory)
>> +{
>> +	struct dma_chan *dma_chan;
>> +	u32 *dma_buf;
>> +	dma_addr_t dma_phys;
>> +	int ret;
>> +	const char *chan_name = dma_to_memory ? "rx" : "tx";
>> +
>> +	dma_chan = dma_request_slave_channel_reason(i2c_dev->dev, chan_name);
>> +	if (IS_ERR(dma_chan))
>> +		return PTR_ERR(dma_chan);
>> +
>> +	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
>> +				     &dma_phys, GFP_KERNEL);
>> +
>> +	if (!dma_buf) {
>> +		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
>> +		ret = -ENOMEM;
>> +		goto scrub;
>> +	}
>> +
>> +	if (dma_to_memory)
>> +		i2c_dev->rx_dma_chan = dma_chan;
>> +	else
>> +		i2c_dev->tx_dma_chan = dma_chan;
>> +
>> +	i2c_dev->dma_buf = dma_buf;
>> +	i2c_dev->dma_phys = dma_phys;
>> +
>> +	return 0;
>> +
>> +scrub:
>> +	dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
>> +			  dma_buf, dma_phys);
> 
> Will dma_free_coherent() handle NULL ptr which will happen when dma_buf allocation fails for the RX init'ing?
> 
>> +	dma_release_channel(dma_chan);
>> +	return ret;
>> +}
>> +
>>  static int tegra_i2c_flush_fifos(struct tegra_i2c_dev *i2c_dev)
>>  {
>>  	unsigned long timeout = jiffies + HZ;
>> @@ -641,25 +745,45 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>>  		goto err;
>>  	}
>>  
>> -	if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
>> -		if (i2c_dev->msg_buf_remaining)
>> -			tegra_i2c_empty_rx_fifo(i2c_dev);
>> -		else
>> -			BUG();
>> -	}
>> +	if (!i2c_dev->is_curr_dma_xfer) {
>> +		if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
>> +			if (i2c_dev->msg_buf_remaining)
>> +				tegra_i2c_empty_rx_fifo(i2c_dev);
>> +			else
>> +				BUG();
>> +		}
>>  
>> -	if (!i2c_dev->msg_read && (status & I2C_INT_TX_FIFO_DATA_REQ)) {
>> -		if (i2c_dev->msg_buf_remaining)
>> -			tegra_i2c_fill_tx_fifo(i2c_dev);
>> -		else
>> -			tegra_i2c_mask_irq(i2c_dev, I2C_INT_TX_FIFO_DATA_REQ);
>> +		if (!i2c_dev->msg_read &&
>> +		   (status & I2C_INT_TX_FIFO_DATA_REQ)) {
>> +			if (i2c_dev->msg_buf_remaining)
>> +				tegra_i2c_fill_tx_fifo(i2c_dev);
>> +			else
>> +				tegra_i2c_mask_irq(i2c_dev,
>> +						   I2C_INT_TX_FIFO_DATA_REQ);
>> +		}
>>  	}
>>  
>>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>>  	if (i2c_dev->is_dvc)
>>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>>  
>> -	if (status & I2C_INT_PACKET_XFER_COMPLETE) {
>> +	if (status & I2C_INT_ALL_PACKETS_XFER_COMPLETE) {
>> +	/*
>> +	 * During message read XFER_COMPLETE interrupt is triggered prior to
>> +	 * DMA complete notification and during message write XFER_COMPLETE
>> +	 * interrupt is triggered after DMA complete notification.
>> +	 * PACKETS_XFER_COMPLETE indicates completion of all bytes of transfer.
>> +	 * so forcing msg_buf_remaining to 0.
>> +	 */
>> +		if (i2c_dev->is_curr_dma_xfer)
>> +			i2c_dev->msg_buf_remaining = 0;
>> +		status |= I2C_INT_PACKET_XFER_COMPLETE;
>> +		i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>> +		if (!i2c_dev->msg_buf_remaining)
>> +			complete(&i2c_dev->msg_complete);
>> +	} else if (status & I2C_INT_PACKET_XFER_COMPLETE) {
>> +		if (i2c_dev->is_curr_dma_xfer)
>> +			i2c_dev->msg_buf_remaining = 0;
>>  		BUG_ON(i2c_dev->msg_buf_remaining);
>>  		complete(&i2c_dev->msg_complete);
>>  	}
>> @@ -668,17 +792,68 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>>  	/* An error occurred, mask all interrupts */
>>  	tegra_i2c_mask_irq(i2c_dev, I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST |
>>  		I2C_INT_PACKET_XFER_COMPLETE | I2C_INT_TX_FIFO_DATA_REQ |
>> -		I2C_INT_RX_FIFO_DATA_REQ);
>> +		I2C_INT_RX_FIFO_DATA_REQ | I2C_INT_ALL_PACKETS_XFER_COMPLETE);
>>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>>  	if (i2c_dev->is_dvc)
>>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>>  
>> +	if (i2c_dev->is_curr_dma_xfer) {
>> +		dmaengine_terminate_all(chan);
>> +		complete(&i2c_dev->dma_complete);
>> +	}
>> +
>>  	complete(&i2c_dev->msg_complete);
>>  done:
>>  	spin_unlock(&i2c_dev->xfer_lock);
>>  	return IRQ_HANDLED;
>>  }
>>  
>> +static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
>> +				       size_t len, int direction)
>> +{
>> +	u32 val, reg;
>> +	u8 dma_burst = 0;
>> +	struct dma_slave_config dma_sconfig;
>> +
>> +	if (i2c_dev->hw->has_mst_fifo)
>> +		reg = I2C_MST_FIFO_CONTROL;
>> +	else
>> +		reg = I2C_FIFO_CONTROL;
>> +	val = i2c_readl(i2c_dev, reg);
>> +
>> +	if (len & 0xF)
>> +		dma_burst = 1;
>> +	else if (len & 0x10)
>> +		dma_burst = 4;
>> +	else
>> +		dma_burst = 8;
> 
> I'm still wondering whether dma_burst < 8 negatively affects transfer efficiency. Could you please clarify?
> 
>> +
>> +	if (direction == DATA_DMA_DIR_TX) {
>> +		if (i2c_dev->hw->has_mst_fifo)
>> +			val |= I2C_MST_FIFO_CONTROL_TX_TRIG(dma_burst);
>> +		else
>> +			val |= I2C_FIFO_CONTROL_TX_TRIG(dma_burst);
>> +	} else {
>> +		if (i2c_dev->hw->has_mst_fifo)
>> +			val |= I2C_MST_FIFO_CONTROL_RX_TRIG(dma_burst);
>> +		else
>> +			val |= I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
>> +	}
>> +	i2c_writel(i2c_dev, val, reg);
>> +
>> +	if (direction == DATA_DMA_DIR_TX) {
>> +		dma_sconfig.dst_addr = i2c_dev->phys_addr + I2C_TX_FIFO;
>> +		dma_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>> +		dma_sconfig.dst_maxburst = dma_burst;
>> +	} else {
>> +		dma_sconfig.src_addr = i2c_dev->phys_addr + I2C_RX_FIFO;
>> +		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>> +		dma_sconfig.src_maxburst = dma_burst;
>> +	}
>> +
>> +	dmaengine_slave_config(chan, &dma_sconfig);
>> +}
>> +
>>  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>>  	struct i2c_msg *msg, enum msg_end_type end_state)
>>  {
>> @@ -688,6 +863,9 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>>  	unsigned long flags;
>>  	u16 xfer_time = 100;
>>  	size_t xfer_size = 0;
>> +	u32 *buffer = 0;
>> +	int ret = 0;
>> +	bool dma = false;
>>  
>>  	if (msg->flags & I2C_M_RD)
>>  		xfer_size = ALIGN(msg->len, BYTES_PER_FIFO_WORD);
>> @@ -698,6 +876,11 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>>  	xfer_time += DIV_ROUND_CLOSEST((xfer_size * 9) * 1000,
>>  					i2c_dev->bus_clk_rate);
>>  
>> +	dma = ((xfer_size > I2C_PIO_MODE_MAX_LEN) &&
>> +		i2c_dev->tx_dma_chan && i2c_dev->rx_dma_chan);
> 
> There is no need to allocate resources for DMA until dma==true, let's postpone channels requesting and dma_buf allocation until they are really needed.

Although DMA channels shall be requested during the probe to maintain suspend-resume order, but dma_buf allocation could be postponed.
Dmitry Osipenko Jan. 26, 2019, 8:42 p.m. UTC | #4
26.01.2019 6:57, Sowjanya Komatineni пишет:
> This patch adds DMA support for Tegra I2C.
> 
> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
> transfer size of the max FIFO depth and DMA mode is used for
> transfer size higher than max FIFO depth to save CPU overhead.
> 
> PIO mode needs full intervention of CPU to fill or empty FIFO's
> and also need to service multiple data requests interrupt for the
> same transaction adding overhead on CPU for large transfers.
> 
> DMA mode is helpful for Large transfers during downloading or
> uploading FW over I2C to some external devices.
> 
> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> 
> ---
>  [V3] : Updated without additional buffer allocation.
>  [V2] : Updated based on V1 review feedback along with code cleanup for
> 	proper implementation of DMA.
> 
>  drivers/i2c/busses/i2c-tegra.c | 341 ++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 316 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
> index 3dcbc9960d9d..452358a77400 100644
> --- a/drivers/i2c/busses/i2c-tegra.c
> +++ b/drivers/i2c/busses/i2c-tegra.c
> @@ -8,6 +8,9 @@
>  
>  #include <linux/clk.h>
>  #include <linux/delay.h>
> +#include <linux/dmaengine.h>
> +#include <linux/dmapool.h>
> +#include <linux/dma-mapping.h>
>  #include <linux/err.h>
>  #include <linux/i2c.h>
>  #include <linux/init.h>
> @@ -45,6 +48,8 @@
>  #define I2C_FIFO_CONTROL_RX_FLUSH		BIT(0)
>  #define I2C_FIFO_CONTROL_TX_TRIG_SHIFT		5
>  #define I2C_FIFO_CONTROL_RX_TRIG_SHIFT		2
> +#define I2C_FIFO_CONTROL_TX_TRIG(x)		(((x) - 1) << 5)
> +#define I2C_FIFO_CONTROL_RX_TRIG(x)		(((x) - 1) << 2)
>  #define I2C_FIFO_STATUS				0x060
>  #define I2C_FIFO_STATUS_TX_MASK			0xF0
>  #define I2C_FIFO_STATUS_TX_SHIFT		4
> @@ -119,6 +124,16 @@
>  /* Packet header size in bytes */
>  #define I2C_PACKET_HEADER_SIZE			12
>  
> +#define DATA_DMA_DIR_TX				(1 << 0)
> +#define DATA_DMA_DIR_RX				(1 << 1)
> +
> +/*
> + * Upto I2C_PIO_MODE_MAX_LEN bytes, controller will use PIO mode,
> + * above this, controller will use DMA to fill FIFO.
> + * MAX PIO len is 20 bytes excluding packet header.
> + */
> +#define I2C_PIO_MODE_MAX_LEN			20
> +
>  /*
>   * msg_end_type: The bus control which need to be send at end of transfer.
>   * @MSG_END_STOP: Send stop pulse at end of transfer.
> @@ -179,6 +194,7 @@ struct tegra_i2c_hw_feature {
>   * @fast_clk: clock reference for fast clock of I2C controller
>   * @rst: reset control for the I2C controller
>   * @base: ioremapped registers cookie
> + * @phys_addr: Physical address of I2C base address to use for DMA configuration
>   * @cont_id: I2C controller ID, used for packet header
>   * @irq: IRQ number of transfer complete interrupt
>   * @irq_disabled: used to track whether or not the interrupt is enabled
> @@ -192,6 +208,14 @@ struct tegra_i2c_hw_feature {
>   * @clk_divisor_non_hs_mode: clock divider for non-high-speed modes
>   * @is_multimaster_mode: track if I2C controller is in multi-master mode
>   * @xfer_lock: lock to serialize transfer submission and processing
> + * @has_dma: indicated if controller supports DMA
> + * @tx_dma_chan: DMA transmit channel
> + * @rx_dma_chan: DMA receive channel
> + * @dma_phys: handle to DMA resources
> + * @dma_buf: pointer to allocated DMA buffer
> + * @dma_buf_size: DMA buffer size
> + * @is_curr_dma_xfer: indicates active DMA transfer
> + * @dma_complete: DMA completion notifier
>   */
>  struct tegra_i2c_dev {
>  	struct device *dev;
> @@ -201,6 +225,7 @@ struct tegra_i2c_dev {
>  	struct clk *fast_clk;
>  	struct reset_control *rst;
>  	void __iomem *base;
> +	phys_addr_t phys_addr;
>  	int cont_id;
>  	int irq;
>  	bool irq_disabled;
> @@ -214,8 +239,18 @@ struct tegra_i2c_dev {
>  	u16 clk_divisor_non_hs_mode;
>  	bool is_multimaster_mode;
>  	spinlock_t xfer_lock;
> +	bool has_dma;
> +	struct dma_chan *tx_dma_chan;
> +	struct dma_chan *rx_dma_chan;
> +	dma_addr_t dma_phys;
> +	u32 *dma_buf;
> +	unsigned int dma_buf_size;
> +	bool is_curr_dma_xfer;
> +	struct completion dma_complete;
>  };
>  
> +static struct dma_chan *chan;
> +
>  static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val,
>  		       unsigned long reg)
>  {
> @@ -282,6 +317,75 @@ static void tegra_i2c_unmask_irq(struct tegra_i2c_dev *i2c_dev, u32 mask)
>  	i2c_writel(i2c_dev, int_mask, I2C_INT_MASK);
>  }
>  
> +static void tegra_i2c_dma_complete(void *args)
> +{
> +	struct tegra_i2c_dev *i2c_dev = args;
> +
> +	complete(&i2c_dev->dma_complete);
> +}
> +
> +static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
> +{
> +	struct dma_async_tx_descriptor *dma_desc;
> +	enum dma_transfer_direction dir;
> +
> +	dev_dbg(i2c_dev->dev, "Starting DMA for length: %zu\n", len);
> +	reinit_completion(&i2c_dev->dma_complete);
> +	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
> +	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
> +					       len, dir, DMA_PREP_INTERRUPT |
> +					       DMA_CTRL_ACK);
> +	if (!dma_desc) {
> +		dev_err(i2c_dev->dev, "Failed to get DMA descriptor\n");
> +		return -EIO;
> +	}
> +
> +	dma_desc->callback = tegra_i2c_dma_complete;
> +	dma_desc->callback_param = i2c_dev;
> +	dmaengine_submit(dma_desc);
> +	dma_async_issue_pending(chan);
> +	return 0;
> +}
> +
> +static int tegra_i2c_init_dma_param(struct tegra_i2c_dev *i2c_dev,
> +				    bool dma_to_memory)
> +{
> +	struct dma_chan *dma_chan;
> +	u32 *dma_buf;
> +	dma_addr_t dma_phys;
> +	int ret;
> +	const char *chan_name = dma_to_memory ? "rx" : "tx";
> +
> +	dma_chan = dma_request_slave_channel_reason(i2c_dev->dev, chan_name);
> +	if (IS_ERR(dma_chan))
> +		return PTR_ERR(dma_chan);
> +
> +	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +				     &dma_phys, GFP_KERNEL);
> +
> +	if (!dma_buf) {
> +		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
> +		ret = -ENOMEM;
> +		goto scrub;
> +	}
> +
> +	if (dma_to_memory)
> +		i2c_dev->rx_dma_chan = dma_chan;
> +	else
> +		i2c_dev->tx_dma_chan = dma_chan;
> +
> +	i2c_dev->dma_buf = dma_buf;
> +	i2c_dev->dma_phys = dma_phys;
> +
> +	return 0;
> +
> +scrub:
> +	dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> +			  dma_buf, dma_phys);
> +	dma_release_channel(dma_chan);
> +	return ret;
> +}
> +
>  static int tegra_i2c_flush_fifos(struct tegra_i2c_dev *i2c_dev)
>  {
>  	unsigned long timeout = jiffies + HZ;
> @@ -641,25 +745,45 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  		goto err;
>  	}
>  
> -	if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_empty_rx_fifo(i2c_dev);
> -		else
> -			BUG();
> -	}
> +	if (!i2c_dev->is_curr_dma_xfer) {
> +		if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_empty_rx_fifo(i2c_dev);
> +			else
> +				BUG();
> +		}
>  
> -	if (!i2c_dev->msg_read && (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> -		if (i2c_dev->msg_buf_remaining)
> -			tegra_i2c_fill_tx_fifo(i2c_dev);
> -		else
> -			tegra_i2c_mask_irq(i2c_dev, I2C_INT_TX_FIFO_DATA_REQ);
> +		if (!i2c_dev->msg_read &&
> +		   (status & I2C_INT_TX_FIFO_DATA_REQ)) {
> +			if (i2c_dev->msg_buf_remaining)
> +				tegra_i2c_fill_tx_fifo(i2c_dev);
> +			else
> +				tegra_i2c_mask_irq(i2c_dev,
> +						   I2C_INT_TX_FIFO_DATA_REQ);
> +		}
>  	}
>  
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> -	if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +	if (status & I2C_INT_ALL_PACKETS_XFER_COMPLETE) {
> +	/*
> +	 * During message read XFER_COMPLETE interrupt is triggered prior to
> +	 * DMA complete notification and during message write XFER_COMPLETE
> +	 * interrupt is triggered after DMA complete notification.
> +	 * PACKETS_XFER_COMPLETE indicates completion of all bytes of transfer.
> +	 * so forcing msg_buf_remaining to 0.
> +	 */
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
> +		status |= I2C_INT_PACKET_XFER_COMPLETE;
> +		i2c_writel(i2c_dev, status, I2C_INT_STATUS);
> +		if (!i2c_dev->msg_buf_remaining)
> +			complete(&i2c_dev->msg_complete);
> +	} else if (status & I2C_INT_PACKET_XFER_COMPLETE) {
> +		if (i2c_dev->is_curr_dma_xfer)
> +			i2c_dev->msg_buf_remaining = 0;
>  		BUG_ON(i2c_dev->msg_buf_remaining);
>  		complete(&i2c_dev->msg_complete);
>  	}
> @@ -668,17 +792,68 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
>  	/* An error occurred, mask all interrupts */
>  	tegra_i2c_mask_irq(i2c_dev, I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST |
>  		I2C_INT_PACKET_XFER_COMPLETE | I2C_INT_TX_FIFO_DATA_REQ |
> -		I2C_INT_RX_FIFO_DATA_REQ);
> +		I2C_INT_RX_FIFO_DATA_REQ | I2C_INT_ALL_PACKETS_XFER_COMPLETE);
>  	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
>  	if (i2c_dev->is_dvc)
>  		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
>  
> +	if (i2c_dev->is_curr_dma_xfer) {
> +		dmaengine_terminate_all(chan);
> +		complete(&i2c_dev->dma_complete);
> +	}
> +
>  	complete(&i2c_dev->msg_complete);
>  done:
>  	spin_unlock(&i2c_dev->xfer_lock);
>  	return IRQ_HANDLED;
>  }
>  
> +static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
> +				       size_t len, int direction)
> +{
> +	u32 val, reg;
> +	u8 dma_burst = 0;
> +	struct dma_slave_config dma_sconfig;
> +
> +	if (i2c_dev->hw->has_mst_fifo)
> +		reg = I2C_MST_FIFO_CONTROL;
> +	else
> +		reg = I2C_FIFO_CONTROL;
> +	val = i2c_readl(i2c_dev, reg);
> +
> +	if (len & 0xF)
> +		dma_burst = 1;
> +	else if (len & 0x10)
> +		dma_burst = 4;
> +	else
> +		dma_burst = 8;
> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_TX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_TX_TRIG(dma_burst);
> +	} else {
> +		if (i2c_dev->hw->has_mst_fifo)
> +			val |= I2C_MST_FIFO_CONTROL_RX_TRIG(dma_burst);
> +		else
> +			val |= I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
> +	}
> +	i2c_writel(i2c_dev, val, reg);
> +
> +	if (direction == DATA_DMA_DIR_TX) {
> +		dma_sconfig.dst_addr = i2c_dev->phys_addr + I2C_TX_FIFO;
> +		dma_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.dst_maxburst = dma_burst;
> +	} else {
> +		dma_sconfig.src_addr = i2c_dev->phys_addr + I2C_RX_FIFO;
> +		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> +		dma_sconfig.src_maxburst = dma_burst;
> +	}
> +
> +	dmaengine_slave_config(chan, &dma_sconfig);
> +}
> +
>  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	struct i2c_msg *msg, enum msg_end_type end_state)
>  {
> @@ -688,6 +863,9 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	unsigned long flags;
>  	u16 xfer_time = 100;
>  	size_t xfer_size = 0;
> +	u32 *buffer = 0;
> +	int ret = 0;
> +	bool dma = false;
>  
>  	if (msg->flags & I2C_M_RD)
>  		xfer_size = ALIGN(msg->len, BYTES_PER_FIFO_WORD);
> @@ -698,6 +876,11 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
>  	xfer_time += DIV_ROUND_CLOSEST((xfer_size * 9) * 1000,
>  					i2c_dev->bus_clk_rate);
>  
> +	dma = ((xfer_size > I2C_PIO_MODE_MAX_LEN) &&
> +		i2c_dev->tx_dma_chan && i2c_dev->rx_dma_chan);

The xfer_size is aligned to 4, does this mean that DMA will transfer more bytes than needed when msg->len isn't multiple of 4?
Dmitry Osipenko Jan. 26, 2019, 9:24 p.m. UTC | #5
26.01.2019 6:57, Sowjanya Komatineni пишет:
> This patch adds DMA support for Tegra I2C.
> 
> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for
> transfer size of the max FIFO depth and DMA mode is used for
> transfer size higher than max FIFO depth to save CPU overhead.
> 
> PIO mode needs full intervention of CPU to fill or empty FIFO's
> and also need to service multiple data requests interrupt for the
> same transaction adding overhead on CPU for large transfers.
> 
> DMA mode is helpful for Large transfers during downloading or
> uploading FW over I2C to some external devices.
> 
> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> 
> ---

Seems there are good news, APB DMA could access IRAM. Quote from TRM: "The source may be DRAM or IRAM, and the destination location could be devices placed on APB Bus; or vice versa". Hence it shall be much more efficient to use IRAM for the DMA buffer, please consider that variant.
Sowjanya Komatineni Jan. 28, 2019, 11:47 p.m. UTC | #6
> > This patch adds DMA support for Tegra I2C.
> > 
> > Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for 
> > transfer size of the max FIFO depth and DMA mode is used for transfer 
> > size higher than max FIFO depth to save CPU overhead.
> > 
> > PIO mode needs full intervention of CPU to fill or empty FIFO's and 
> > also need to service multiple data requests interrupt for the same 
> > transaction adding overhead on CPU for large transfers.
> > 
> > DMA mode is helpful for Large transfers during downloading or 
> > uploading FW over I2C to some external devices.
> > 
> > Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> > 
> > ---
>
> Seems there are good news, APB DMA could access IRAM. Quote from TRM: "The source may be DRAM or IRAM, and the destination location could be devices placed on > APB Bus; or vice versa". Hence it shall be much more efficient to use IRAM for the DMA buffer, please consider that variant.
>

Regarding your request for Write Combined memory, it doesn’t guarantee the order.

Main reason for adding DMA mode support to Tegra I2C driver is not for actual transfer performance but for not having delayed I2C transfer mainly in case of CPU fully loaded and I2C transfer during that time (FW upload/download, Sensor programming..) using PIO mode introduces delay b/w bytes of transfer as it depends on CPU scheduling and interrupt priority which impacts on serving interrupts and filling FIFO for transfer to happen.

Below are the cases where we see the need for DMA mode for I2C transfers (mainly during the use cases where CPU is fully loaded) 
- Some devices need FW update over I2C and DMA mode helps in such large transfers on I2C reducing CPU intervention mainly when CPU is fully loaded.
- Some devices have internal timeout for no data activity on I2C bus after the start command (time restriction b/w data to data within single transaction) to avoid bus hang and they force to stop when timeout happens. with PIO mode in those cases, depending on CPU load, there could be more delay in serving interrupt for data requests and sending the data causing the device to trigger its timeout and transaction stops.
- Helps for reading data from I2C sensors with better BW.
Sowjanya Komatineni Jan. 29, 2019, 2:02 a.m. UTC | #7
> > > This patch adds DMA support for Tegra I2C.
> > > 
> > > Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for 
> > > transfer size of the max FIFO depth and DMA mode is used for 
> > > transfer size higher than max FIFO depth to save CPU overhead.
> > > 
> > > PIO mode needs full intervention of CPU to fill or empty FIFO's and 
> > > also need to service multiple data requests interrupt for the same 
> > > transaction adding overhead on CPU for large transfers.
> > > 
> > > DMA mode is helpful for Large transfers during downloading or 
> > > uploading FW over I2C to some external devices.
> > > 
> > > Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
> > > 
> > > ---
> >
> > Seems there are good news, APB DMA could access IRAM. Quote from TRM: "The source may be DRAM or IRAM, and the destination location could be devices placed on > APB Bus; or vice versa". Hence it shall be much more efficient to use IRAM for the DMA buffer, please consider that variant.
> >
>
> Regarding your request for Write Combined memory, it doesn’t guarantee the order.
>
> Main reason for adding DMA mode support to Tegra I2C driver is not for actual transfer performance but for not having delayed I2C transfer mainly in case of CPU fully loaded and I2C transfer during that time (FW upload/download, Sensor programming..) using PIO > mode introduces delay b/w bytes of transfer as it depends on CPU scheduling and interrupt priority which impacts on serving interrupts and filling FIFO for transfer to happen.
>
> Below are the cases where we see the need for DMA mode for I2C transfers (mainly during the use cases where CPU is fully loaded)
> - Some devices need FW update over I2C and DMA mode helps in such large transfers on I2C reducing CPU intervention mainly when CPU is fully loaded.
> - Some devices have internal timeout for no data activity on I2C bus after the start command (time restriction b/w data to data within single transaction) to avoid bus hang and they force to stop when timeout happens. with PIO mode in those cases, depending on CPU load, there could be more delay in serving interrupt for data requests and sending the data causing the device to trigger its timeout and transaction stops.
> - Helps for reading data from I2C sensors with better BW.

Found from documentation default attribute is DMA_ATTR_WRITE_BARRIER which forces all pending writes to complete preventing race b/w completion notification and data DMA.
Will use write-combined memory allocation for DMA buffer in revised patch..

Thanks
sowjanya
Dmitry Osipenko Jan. 29, 2019, 3:11 a.m. UTC | #8
29.01.2019 5:02, Sowjanya Komatineni пишет:
> 
> 
>>>> This patch adds DMA support for Tegra I2C.
>>>>
>>>> Tegra I2C TX and RX FIFO depth is 8 words. PIO mode is used for 
>>>> transfer size of the max FIFO depth and DMA mode is used for 
>>>> transfer size higher than max FIFO depth to save CPU overhead.
>>>>
>>>> PIO mode needs full intervention of CPU to fill or empty FIFO's and 
>>>> also need to service multiple data requests interrupt for the same 
>>>> transaction adding overhead on CPU for large transfers.
>>>>
>>>> DMA mode is helpful for Large transfers during downloading or 
>>>> uploading FW over I2C to some external devices.
>>>>
>>>> Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
>>>>
>>>> ---
>>>
>>> Seems there are good news, APB DMA could access IRAM. Quote from TRM: "The source may be DRAM or IRAM, and the destination location could be devices placed on > APB Bus; or vice versa". Hence it shall be much more efficient to use IRAM for the DMA buffer, please consider that variant.
>>>
>>
>> Regarding your request for Write Combined memory, it doesn’t guarantee the order.

The memory-access order doesn't matter at all in your case, all you need to care about is that actual data is read from dma-buffer and that caches are flushed after writing is done.

>> Main reason for adding DMA mode support to Tegra I2C driver is not for actual transfer performance but for not having delayed I2C transfer mainly in case of CPU fully loaded and I2C transfer during that time (FW upload/download, Sensor programming..) using PIO > mode introduces delay b/w bytes of transfer as it depends on CPU scheduling and interrupt priority which impacts on serving interrupts and filling FIFO for transfer to happen.
>>
>> Below are the cases where we see the need for DMA mode for I2C transfers (mainly during the use cases where CPU is fully loaded)
>> - Some devices need FW update over I2C and DMA mode helps in such large transfers on I2C reducing CPU intervention mainly when CPU is fully loaded.
>> - Some devices have internal timeout for no data activity on I2C bus after the start command (time restriction b/w data to data within single transaction) to avoid bus hang and they force to stop when timeout happens. with PIO mode in those cases, depending on CPU load, there could be more delay in serving interrupt for data requests and sending the data causing the device to trigger its timeout and transaction stops.
>> - Helps for reading data from I2C sensors with better BW.

Okay, since the efficiency between transfers is not the concern in your case, it doesn't really matter much where dma-buffer is located.

> Found from documentation default attribute is DMA_ATTR_WRITE_BARRIER which forces all pending writes to complete preventing race b/w completion notification and data DMA.
> Will use write-combined memory allocation for DMA buffer in revised patch..

Actually mapping is write-combined on ARM for the coherent DMA allocation with CONFIG_ARM_DMA_MEM_BUFFERABLE=y (enabled by default). I just falsely assumed that it could be writeback. Hence please keep the coherent allocation as-is. And you don't need to care about DMA_ATTR_WRITE_BARRIER, apparently it is only relevant for PCI on IA64.

Patch
diff mbox series

diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
index 3dcbc9960d9d..452358a77400 100644
--- a/drivers/i2c/busses/i2c-tegra.c
+++ b/drivers/i2c/busses/i2c-tegra.c
@@ -8,6 +8,9 @@ 
 
 #include <linux/clk.h>
 #include <linux/delay.h>
+#include <linux/dmaengine.h>
+#include <linux/dmapool.h>
+#include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/i2c.h>
 #include <linux/init.h>
@@ -45,6 +48,8 @@ 
 #define I2C_FIFO_CONTROL_RX_FLUSH		BIT(0)
 #define I2C_FIFO_CONTROL_TX_TRIG_SHIFT		5
 #define I2C_FIFO_CONTROL_RX_TRIG_SHIFT		2
+#define I2C_FIFO_CONTROL_TX_TRIG(x)		(((x) - 1) << 5)
+#define I2C_FIFO_CONTROL_RX_TRIG(x)		(((x) - 1) << 2)
 #define I2C_FIFO_STATUS				0x060
 #define I2C_FIFO_STATUS_TX_MASK			0xF0
 #define I2C_FIFO_STATUS_TX_SHIFT		4
@@ -119,6 +124,16 @@ 
 /* Packet header size in bytes */
 #define I2C_PACKET_HEADER_SIZE			12
 
+#define DATA_DMA_DIR_TX				(1 << 0)
+#define DATA_DMA_DIR_RX				(1 << 1)
+
+/*
+ * Upto I2C_PIO_MODE_MAX_LEN bytes, controller will use PIO mode,
+ * above this, controller will use DMA to fill FIFO.
+ * MAX PIO len is 20 bytes excluding packet header.
+ */
+#define I2C_PIO_MODE_MAX_LEN			20
+
 /*
  * msg_end_type: The bus control which need to be send at end of transfer.
  * @MSG_END_STOP: Send stop pulse at end of transfer.
@@ -179,6 +194,7 @@  struct tegra_i2c_hw_feature {
  * @fast_clk: clock reference for fast clock of I2C controller
  * @rst: reset control for the I2C controller
  * @base: ioremapped registers cookie
+ * @phys_addr: Physical address of I2C base address to use for DMA configuration
  * @cont_id: I2C controller ID, used for packet header
  * @irq: IRQ number of transfer complete interrupt
  * @irq_disabled: used to track whether or not the interrupt is enabled
@@ -192,6 +208,14 @@  struct tegra_i2c_hw_feature {
  * @clk_divisor_non_hs_mode: clock divider for non-high-speed modes
  * @is_multimaster_mode: track if I2C controller is in multi-master mode
  * @xfer_lock: lock to serialize transfer submission and processing
+ * @has_dma: indicated if controller supports DMA
+ * @tx_dma_chan: DMA transmit channel
+ * @rx_dma_chan: DMA receive channel
+ * @dma_phys: handle to DMA resources
+ * @dma_buf: pointer to allocated DMA buffer
+ * @dma_buf_size: DMA buffer size
+ * @is_curr_dma_xfer: indicates active DMA transfer
+ * @dma_complete: DMA completion notifier
  */
 struct tegra_i2c_dev {
 	struct device *dev;
@@ -201,6 +225,7 @@  struct tegra_i2c_dev {
 	struct clk *fast_clk;
 	struct reset_control *rst;
 	void __iomem *base;
+	phys_addr_t phys_addr;
 	int cont_id;
 	int irq;
 	bool irq_disabled;
@@ -214,8 +239,18 @@  struct tegra_i2c_dev {
 	u16 clk_divisor_non_hs_mode;
 	bool is_multimaster_mode;
 	spinlock_t xfer_lock;
+	bool has_dma;
+	struct dma_chan *tx_dma_chan;
+	struct dma_chan *rx_dma_chan;
+	dma_addr_t dma_phys;
+	u32 *dma_buf;
+	unsigned int dma_buf_size;
+	bool is_curr_dma_xfer;
+	struct completion dma_complete;
 };
 
+static struct dma_chan *chan;
+
 static void dvc_writel(struct tegra_i2c_dev *i2c_dev, u32 val,
 		       unsigned long reg)
 {
@@ -282,6 +317,75 @@  static void tegra_i2c_unmask_irq(struct tegra_i2c_dev *i2c_dev, u32 mask)
 	i2c_writel(i2c_dev, int_mask, I2C_INT_MASK);
 }
 
+static void tegra_i2c_dma_complete(void *args)
+{
+	struct tegra_i2c_dev *i2c_dev = args;
+
+	complete(&i2c_dev->dma_complete);
+}
+
+static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
+{
+	struct dma_async_tx_descriptor *dma_desc;
+	enum dma_transfer_direction dir;
+
+	dev_dbg(i2c_dev->dev, "Starting DMA for length: %zu\n", len);
+	reinit_completion(&i2c_dev->dma_complete);
+	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
+	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
+					       len, dir, DMA_PREP_INTERRUPT |
+					       DMA_CTRL_ACK);
+	if (!dma_desc) {
+		dev_err(i2c_dev->dev, "Failed to get DMA descriptor\n");
+		return -EIO;
+	}
+
+	dma_desc->callback = tegra_i2c_dma_complete;
+	dma_desc->callback_param = i2c_dev;
+	dmaengine_submit(dma_desc);
+	dma_async_issue_pending(chan);
+	return 0;
+}
+
+static int tegra_i2c_init_dma_param(struct tegra_i2c_dev *i2c_dev,
+				    bool dma_to_memory)
+{
+	struct dma_chan *dma_chan;
+	u32 *dma_buf;
+	dma_addr_t dma_phys;
+	int ret;
+	const char *chan_name = dma_to_memory ? "rx" : "tx";
+
+	dma_chan = dma_request_slave_channel_reason(i2c_dev->dev, chan_name);
+	if (IS_ERR(dma_chan))
+		return PTR_ERR(dma_chan);
+
+	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
+				     &dma_phys, GFP_KERNEL);
+
+	if (!dma_buf) {
+		dev_err(i2c_dev->dev, "Failed to allocate the DMA buffer\n");
+		ret = -ENOMEM;
+		goto scrub;
+	}
+
+	if (dma_to_memory)
+		i2c_dev->rx_dma_chan = dma_chan;
+	else
+		i2c_dev->tx_dma_chan = dma_chan;
+
+	i2c_dev->dma_buf = dma_buf;
+	i2c_dev->dma_phys = dma_phys;
+
+	return 0;
+
+scrub:
+	dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
+			  dma_buf, dma_phys);
+	dma_release_channel(dma_chan);
+	return ret;
+}
+
 static int tegra_i2c_flush_fifos(struct tegra_i2c_dev *i2c_dev)
 {
 	unsigned long timeout = jiffies + HZ;
@@ -641,25 +745,45 @@  static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
 		goto err;
 	}
 
-	if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
-		if (i2c_dev->msg_buf_remaining)
-			tegra_i2c_empty_rx_fifo(i2c_dev);
-		else
-			BUG();
-	}
+	if (!i2c_dev->is_curr_dma_xfer) {
+		if (i2c_dev->msg_read && (status & I2C_INT_RX_FIFO_DATA_REQ)) {
+			if (i2c_dev->msg_buf_remaining)
+				tegra_i2c_empty_rx_fifo(i2c_dev);
+			else
+				BUG();
+		}
 
-	if (!i2c_dev->msg_read && (status & I2C_INT_TX_FIFO_DATA_REQ)) {
-		if (i2c_dev->msg_buf_remaining)
-			tegra_i2c_fill_tx_fifo(i2c_dev);
-		else
-			tegra_i2c_mask_irq(i2c_dev, I2C_INT_TX_FIFO_DATA_REQ);
+		if (!i2c_dev->msg_read &&
+		   (status & I2C_INT_TX_FIFO_DATA_REQ)) {
+			if (i2c_dev->msg_buf_remaining)
+				tegra_i2c_fill_tx_fifo(i2c_dev);
+			else
+				tegra_i2c_mask_irq(i2c_dev,
+						   I2C_INT_TX_FIFO_DATA_REQ);
+		}
 	}
 
 	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
 	if (i2c_dev->is_dvc)
 		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
 
-	if (status & I2C_INT_PACKET_XFER_COMPLETE) {
+	if (status & I2C_INT_ALL_PACKETS_XFER_COMPLETE) {
+	/*
+	 * During message read XFER_COMPLETE interrupt is triggered prior to
+	 * DMA complete notification and during message write XFER_COMPLETE
+	 * interrupt is triggered after DMA complete notification.
+	 * PACKETS_XFER_COMPLETE indicates completion of all bytes of transfer.
+	 * so forcing msg_buf_remaining to 0.
+	 */
+		if (i2c_dev->is_curr_dma_xfer)
+			i2c_dev->msg_buf_remaining = 0;
+		status |= I2C_INT_PACKET_XFER_COMPLETE;
+		i2c_writel(i2c_dev, status, I2C_INT_STATUS);
+		if (!i2c_dev->msg_buf_remaining)
+			complete(&i2c_dev->msg_complete);
+	} else if (status & I2C_INT_PACKET_XFER_COMPLETE) {
+		if (i2c_dev->is_curr_dma_xfer)
+			i2c_dev->msg_buf_remaining = 0;
 		BUG_ON(i2c_dev->msg_buf_remaining);
 		complete(&i2c_dev->msg_complete);
 	}
@@ -668,17 +792,68 @@  static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
 	/* An error occurred, mask all interrupts */
 	tegra_i2c_mask_irq(i2c_dev, I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST |
 		I2C_INT_PACKET_XFER_COMPLETE | I2C_INT_TX_FIFO_DATA_REQ |
-		I2C_INT_RX_FIFO_DATA_REQ);
+		I2C_INT_RX_FIFO_DATA_REQ | I2C_INT_ALL_PACKETS_XFER_COMPLETE);
 	i2c_writel(i2c_dev, status, I2C_INT_STATUS);
 	if (i2c_dev->is_dvc)
 		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
 
+	if (i2c_dev->is_curr_dma_xfer) {
+		dmaengine_terminate_all(chan);
+		complete(&i2c_dev->dma_complete);
+	}
+
 	complete(&i2c_dev->msg_complete);
 done:
 	spin_unlock(&i2c_dev->xfer_lock);
 	return IRQ_HANDLED;
 }
 
+static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
+				       size_t len, int direction)
+{
+	u32 val, reg;
+	u8 dma_burst = 0;
+	struct dma_slave_config dma_sconfig;
+
+	if (i2c_dev->hw->has_mst_fifo)
+		reg = I2C_MST_FIFO_CONTROL;
+	else
+		reg = I2C_FIFO_CONTROL;
+	val = i2c_readl(i2c_dev, reg);
+
+	if (len & 0xF)
+		dma_burst = 1;
+	else if (len & 0x10)
+		dma_burst = 4;
+	else
+		dma_burst = 8;
+
+	if (direction == DATA_DMA_DIR_TX) {
+		if (i2c_dev->hw->has_mst_fifo)
+			val |= I2C_MST_FIFO_CONTROL_TX_TRIG(dma_burst);
+		else
+			val |= I2C_FIFO_CONTROL_TX_TRIG(dma_burst);
+	} else {
+		if (i2c_dev->hw->has_mst_fifo)
+			val |= I2C_MST_FIFO_CONTROL_RX_TRIG(dma_burst);
+		else
+			val |= I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
+	}
+	i2c_writel(i2c_dev, val, reg);
+
+	if (direction == DATA_DMA_DIR_TX) {
+		dma_sconfig.dst_addr = i2c_dev->phys_addr + I2C_TX_FIFO;
+		dma_sconfig.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
+		dma_sconfig.dst_maxburst = dma_burst;
+	} else {
+		dma_sconfig.src_addr = i2c_dev->phys_addr + I2C_RX_FIFO;
+		dma_sconfig.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
+		dma_sconfig.src_maxburst = dma_burst;
+	}
+
+	dmaengine_slave_config(chan, &dma_sconfig);
+}
+
 static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 	struct i2c_msg *msg, enum msg_end_type end_state)
 {
@@ -688,6 +863,9 @@  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 	unsigned long flags;
 	u16 xfer_time = 100;
 	size_t xfer_size = 0;
+	u32 *buffer = 0;
+	int ret = 0;
+	bool dma = false;
 
 	if (msg->flags & I2C_M_RD)
 		xfer_size = ALIGN(msg->len, BYTES_PER_FIFO_WORD);
@@ -698,6 +876,11 @@  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 	xfer_time += DIV_ROUND_CLOSEST((xfer_size * 9) * 1000,
 					i2c_dev->bus_clk_rate);
 
+	dma = ((xfer_size > I2C_PIO_MODE_MAX_LEN) &&
+		i2c_dev->tx_dma_chan && i2c_dev->rx_dma_chan);
+
+	i2c_dev->is_curr_dma_xfer = dma;
+
 	tegra_i2c_flush_fifos(i2c_dev);
 
 	i2c_dev->msg_buf = msg->buf;
@@ -711,14 +894,50 @@  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 	int_mask = I2C_INT_NO_ACK | I2C_INT_ARBITRATION_LOST;
 	tegra_i2c_unmask_irq(i2c_dev, int_mask);
 
+	if (dma) {
+		if (i2c_dev->msg_read) {
+			chan = i2c_dev->rx_dma_chan;
+			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
+						   DATA_DMA_DIR_RX);
+			/* make the dma buffer to read by dma */
+			dma_sync_single_for_device(i2c_dev->dev,
+						   i2c_dev->dma_phys,
+						   i2c_dev->dma_buf_size,
+						   DMA_FROM_DEVICE);
+			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
+			if (ret < 0) {
+				dev_err(i2c_dev->dev,
+					"Starting RX DMA failed, err %d\n",
+					ret);
+				goto exit;
+			}
+		} else {
+			chan = i2c_dev->tx_dma_chan;
+			tegra_i2c_config_fifo_trig(i2c_dev, xfer_size,
+						   DATA_DMA_DIR_TX);
+			/* Make the dma buffer to read by cpu */
+			dma_sync_single_for_cpu(i2c_dev->dev,
+						i2c_dev->dma_phys,
+						i2c_dev->dma_buf_size,
+						DMA_TO_DEVICE);
+			buffer = (u32 *)i2c_dev->dma_buf;
+		}
+	}
+
 	packet_header = (0 << PACKET_HEADER0_HEADER_SIZE_SHIFT) |
 			PACKET_HEADER0_PROTOCOL_I2C |
 			(i2c_dev->cont_id << PACKET_HEADER0_CONT_ID_SHIFT) |
 			(1 << PACKET_HEADER0_PACKET_ID_SHIFT);
-	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
+	if (dma && !i2c_dev->msg_read)
+		*buffer++ = packet_header;
+	else
+		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
 
 	packet_header = msg->len - 1;
-	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
+	if (dma && !i2c_dev->msg_read)
+		*buffer++ = packet_header;
+	else
+		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
 
 	packet_header = I2C_HEADER_IE_ENABLE;
 	if (end_state == MSG_END_CONTINUE)
@@ -735,29 +954,85 @@  static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 		packet_header |= I2C_HEADER_CONT_ON_NAK;
 	if (msg->flags & I2C_M_RD)
 		packet_header |= I2C_HEADER_READ;
-	i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
-
-	if (!(msg->flags & I2C_M_RD))
-		tegra_i2c_fill_tx_fifo(i2c_dev);
-
+	if (dma && !i2c_dev->msg_read)
+		*buffer++ = packet_header;
+	else
+		i2c_writel(i2c_dev, packet_header, I2C_TX_FIFO);
+
+	if (!i2c_dev->msg_read) {
+		if (dma) {
+			memcpy(buffer, msg->buf, msg->len);
+			/* make the dma buffer to read by dma */
+			dma_sync_single_for_device(i2c_dev->dev,
+						   i2c_dev->dma_phys,
+						   i2c_dev->dma_buf_size,
+						   DMA_TO_DEVICE);
+			ret = tegra_i2c_dma_submit(i2c_dev, xfer_size);
+			if (ret < 0) {
+				dev_err(i2c_dev->dev,
+					"Starting TX DMA failed, err %d\n",
+					ret);
+				goto exit;
+			}
+		} else {
+			tegra_i2c_fill_tx_fifo(i2c_dev);
+		}
+	}
 	if (i2c_dev->hw->has_per_pkt_xfer_complete_irq)
 		int_mask |= I2C_INT_PACKET_XFER_COMPLETE;
-	if (msg->flags & I2C_M_RD)
-		int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
-	else if (i2c_dev->msg_buf_remaining)
-		int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
+	if (dma) {
+		int_mask |= I2C_INT_ALL_PACKETS_XFER_COMPLETE;
+	} else {
+		if (msg->flags & I2C_M_RD)
+			int_mask |= I2C_INT_RX_FIFO_DATA_REQ;
+		else if (i2c_dev->msg_buf_remaining)
+			int_mask |= I2C_INT_TX_FIFO_DATA_REQ;
+	}
 
 	tegra_i2c_unmask_irq(i2c_dev, int_mask);
-	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
 	dev_dbg(i2c_dev->dev, "unmasked irq: %02x\n",
 		i2c_readl(i2c_dev, I2C_INT_MASK));
 
+exit:
+	spin_unlock_irqrestore(&i2c_dev->xfer_lock, flags);
+	if (ret)
+		return ret;
+
+	if (dma) {
+		time_left = wait_for_completion_timeout(
+						&i2c_dev->dma_complete,
+						TEGRA_I2C_TIMEOUT(xfer_time));
+
+		if (time_left == 0) {
+			dev_err(i2c_dev->dev, "DMA transfer timeout\n");
+			dmaengine_terminate_all(chan);
+			tegra_i2c_init(i2c_dev);
+			return -ETIMEDOUT;
+		}
+
+		if (i2c_dev->msg_read) {
+			if (likely(i2c_dev->msg_err == I2C_ERR_NONE)) {
+				dma_sync_single_for_cpu(i2c_dev->dev,
+							i2c_dev->dma_phys,
+							i2c_dev->dma_buf_size,
+							DMA_FROM_DEVICE);
+
+				memcpy(i2c_dev->msg_buf, i2c_dev->dma_buf,
+					msg->len);
+			}
+		}
+	}
+
 	time_left = wait_for_completion_timeout(&i2c_dev->msg_complete,
 						TEGRA_I2C_TIMEOUT(xfer_time));
 	tegra_i2c_mask_irq(i2c_dev, int_mask);
 
 	if (time_left == 0) {
 		dev_err(i2c_dev->dev, "i2c transfer timed out\n");
+		if (dma) {
+			dmaengine_terminate_all(chan);
+			complete(&i2c_dev->dma_complete);
+		}
 
 		tegra_i2c_init(i2c_dev);
 		return -ETIMEDOUT;
@@ -835,6 +1110,8 @@  static void tegra_i2c_parse_dt(struct tegra_i2c_dev *i2c_dev)
 
 	i2c_dev->is_multimaster_mode = of_property_read_bool(np,
 			"multi-master");
+
+	i2c_dev->has_dma = of_property_read_bool(np, "dmas");
 }
 
 static const struct i2c_algorithm tegra_i2c_algo = {
@@ -947,11 +1224,13 @@  static int tegra_i2c_probe(struct platform_device *pdev)
 	struct clk *div_clk;
 	struct clk *fast_clk;
 	void __iomem *base;
+	phys_addr_t phys_addr;
 	int irq;
 	int ret = 0;
 	int clk_multiplier = I2C_CLK_MULTIPLIER_STD_FAST_MODE;
 
 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	phys_addr = res->start;
 	base = devm_ioremap_resource(&pdev->dev, res);
 	if (IS_ERR(base))
 		return PTR_ERR(base);
@@ -974,12 +1253,14 @@  static int tegra_i2c_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	i2c_dev->base = base;
+	i2c_dev->phys_addr = phys_addr;
 	i2c_dev->div_clk = div_clk;
 	i2c_dev->adapter.algo = &tegra_i2c_algo;
 	i2c_dev->adapter.quirks = &tegra_i2c_quirks;
 	i2c_dev->irq = irq;
 	i2c_dev->cont_id = pdev->id;
 	i2c_dev->dev = &pdev->dev;
+	i2c_dev->dma_buf_size = i2c_dev->adapter.quirks->max_write_len;
 
 	i2c_dev->rst = devm_reset_control_get_exclusive(&pdev->dev, "i2c");
 	if (IS_ERR(i2c_dev->rst)) {
@@ -994,6 +1275,7 @@  static int tegra_i2c_probe(struct platform_device *pdev)
 						  "nvidia,tegra20-i2c-dvc");
 	init_completion(&i2c_dev->msg_complete);
 	spin_lock_init(&i2c_dev->xfer_lock);
+	init_completion(&i2c_dev->dma_complete);
 
 	if (!i2c_dev->hw->has_single_clk_source) {
 		fast_clk = devm_clk_get(&pdev->dev, "fast-clk");
@@ -1053,6 +1335,15 @@  static int tegra_i2c_probe(struct platform_device *pdev)
 		}
 	}
 
+	if (i2c_dev->has_dma) {
+		ret = tegra_i2c_init_dma_param(i2c_dev, true);
+		if (ret == -EPROBE_DEFER)
+			goto disable_div_clk;
+		ret = tegra_i2c_init_dma_param(i2c_dev, false);
+		if (ret == -EPROBE_DEFER)
+			goto disable_div_clk;
+	}
+
 	ret = tegra_i2c_init(i2c_dev);
 	if (ret) {
 		dev_err(&pdev->dev, "Failed to initialize i2c controller\n");