diff mbox

[1/3] i2c-mv64xxx: Add I2C Transaction Generator support

Message ID 1373898278-4805-2-git-send-email-gregory.clement@free-electrons.com
State Superseded
Headers show

Commit Message

Gregory CLEMENT July 15, 2013, 2:24 p.m. UTC
The I2C Transaction Generator offloads CPU from managing I2C
transfer step by step.

This feature is currently only available on Armada XP, so usage of
this mechanism is activated through device tree.

Based on the work of Piotr Ziecik and rewrote to use the new way of
handling multiples i2c messages.

Signed-off-by: Piotr Ziecik <kosmo@semihalf.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com>
---
 drivers/i2c/busses/i2c-mv64xxx.c | 207 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 196 insertions(+), 11 deletions(-)

Comments

Maxime Ripard July 16, 2013, 8:05 a.m. UTC | #1
Hi Gregory,

On Mon, Jul 15, 2013 at 04:24:36PM +0200, Gregory CLEMENT wrote:
> The I2C Transaction Generator offloads CPU from managing I2C
> transfer step by step.
> 
> This feature is currently only available on Armada XP, so usage of
> this mechanism is activated through device tree.
> 
> Based on the work of Piotr Ziecik and rewrote to use the new way of
> handling multiples i2c messages.
> 
> Signed-off-by: Piotr Ziecik <kosmo@semihalf.com>
> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com>
> ---
>  drivers/i2c/busses/i2c-mv64xxx.c | 207 ++++++++++++++++++++++++++++++++++++---
>  1 file changed, 196 insertions(+), 11 deletions(-)

[...]

> +	/*
> +	 * For controllers embedded in new SoCs activate the
> +	 * Transaction Generator support.
> +	 */
> +	if (of_device_is_compatible(np, "marvell,mv78230-i2c"))
> +		drv_data->offload_enabled = true;
> +

Do you have a reason for not adding it to the match table? I mean, you
will introduce a new compatible here, but if that compatible is used
alone, won't probe the driver? That doesn't seem very right to me.

Also, you should probably add it to the bindings documentation.

Maxime
Gregory CLEMENT Aug. 6, 2013, 12:05 p.m. UTC | #2
On 16/07/2013 10:05, Maxime Ripard wrote:
> Hi Gregory,
> 
> On Mon, Jul 15, 2013 at 04:24:36PM +0200, Gregory CLEMENT wrote:
>> The I2C Transaction Generator offloads CPU from managing I2C transfer step by step.
>> 
>> This feature is currently only available on Armada XP, so usage of this mechanism is activated through device tree.
>> 
>> Based on the work of Piotr Ziecik and rewrote to use the new way of handling multiples i2c messages.
>> 
>> Signed-off-by: Piotr Ziecik <kosmo@semihalf.com> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> --- drivers/i2c/busses/i2c-mv64xxx.c | 207
>> ++++++++++++++++++++++++++++++++++++--- 1 file changed, 196 insertions(+), 11 deletions(-)
> 
> [...]
> 
>> +	/* +	 * For controllers embedded in new SoCs activate the +	 * Transaction Generator support. +	 */ +	if (of_device_is_compatible(np, "marvell,mv78230-i2c")) +
>> drv_data->offload_enabled = true; +
> 
> Do you have a reason for not adding it to the match table? I mean, you will introduce a new compatible here, but if that compatible is used alone, won't probe the driver? That doesn't
> seem very right to me.

But we shouldn't use it alone: we should always use:
compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";

From my point of view using  "marvell,mv78230-i2c" alone is an error.

Wolfram what is your opinion on it?


> 
> Also, you should probably add it to the bindings documentation.
See the patch 3 for the bindings documentation.

> 
> Maxime
>
Maxime Ripard Aug. 6, 2013, 4:21 p.m. UTC | #3
On Tue, Aug 06, 2013 at 02:05:24PM +0200, Gregory CLEMENT wrote:
> On 16/07/2013 10:05, Maxime Ripard wrote:
> > Hi Gregory,
> > 
> > On Mon, Jul 15, 2013 at 04:24:36PM +0200, Gregory CLEMENT wrote:
> >> The I2C Transaction Generator offloads CPU from managing I2C transfer step by step.
> >> 
> >> This feature is currently only available on Armada XP, so usage of this mechanism is activated through device tree.
> >> 
> >> Based on the work of Piotr Ziecik and rewrote to use the new way of handling multiples i2c messages.
> >> 
> >> Signed-off-by: Piotr Ziecik <kosmo@semihalf.com> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> --- drivers/i2c/busses/i2c-mv64xxx.c | 207
> >> ++++++++++++++++++++++++++++++++++++--- 1 file changed, 196 insertions(+), 11 deletions(-)
> > 
> > [...]
> > 
> >> +	/* +	 * For controllers embedded in new SoCs activate the +	 * Transaction Generator support. +	 */ +	if (of_device_is_compatible(np, "marvell,mv78230-i2c")) +
> >> drv_data->offload_enabled = true; +
> > 
> > Do you have a reason for not adding it to the match table? I mean, you will introduce a new compatible here, but if that compatible is used alone, won't probe the driver? That doesn't
> > seem very right to me.
> 
> But we shouldn't use it alone: we should always use:
> compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";
> 
> From my point of view using  "marvell,mv78230-i2c" alone is an error.

Why is that? If the I2C controller is a new IP with additional features,
it should have a full compatible of its own, doesn't it?
Thomas Petazzoni Aug. 6, 2013, 5:55 p.m. UTC | #4
Dear Gregory CLEMENT,

On Tue, 06 Aug 2013 14:05:24 +0200, Gregory CLEMENT wrote:

> But we shouldn't use it alone: we should always use:
> compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";
> 
> From my point of view using  "marvell,mv78230-i2c" alone is an error.

I disagree. I believe the driver should support a Device Tree file that
uses "marvell,mv78230-i2c" alone.

Best regards,

Thomas
Wolfram Sang Aug. 7, 2013, 2:35 p.m. UTC | #5
> But we shouldn't use it alone: we should always use:
> compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";
> 
> From my point of view using  "marvell,mv78230-i2c" alone is an error.
> 
> Wolfram what is your opinion on it?

It is not strictly an error, but risky. If you use an older Kernel
version (or other OS) which only offers "mv64xxx" you will have no
match. Although the driver theoretically could have basic support for
all mv64xxx variants skipping all additional features of later IP
revisions.
Jason Cooper Aug. 7, 2013, 3:57 p.m. UTC | #6
On Wed, Aug 07, 2013 at 04:35:46PM +0200, Wolfram Sang wrote:
> 
> > But we shouldn't use it alone: we should always use:
> > compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";
> > 
> > From my point of view using  "marvell,mv78230-i2c" alone is an error.
> > 
> > Wolfram what is your opinion on it?
> 
> It is not strictly an error, but risky. If you use an older Kernel
> version (or other OS) which only offers "mv64xxx" you will have no
> match. Although the driver theoretically could have basic support for
> all mv64xxx variants skipping all additional features of later IP
> revisions.

I agree here.  The driver is advertising what IP blocks it can handle,
so it makes sense to add both strings since it can handle both.

thx,

Jason.
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Gregory CLEMENT Aug. 8, 2013, 3:30 p.m. UTC | #7
On 07/08/2013 17:57, Jason Cooper wrote:
> On Wed, Aug 07, 2013 at 04:35:46PM +0200, Wolfram Sang wrote:
>>
>>> But we shouldn't use it alone: we should always use:
>>> compatible = "marvell,mv78230-i2c", "marvell,mv64xxx-i2c";
>>>
>>> From my point of view using  "marvell,mv78230-i2c" alone is an error.
>>>
>>> Wolfram what is your opinion on it?
>>
>> It is not strictly an error, but risky. If you use an older Kernel
>> version (or other OS) which only offers "mv64xxx" you will have no
>> match. Although the driver theoretically could have basic support for
>> all mv64xxx variants skipping all additional features of later IP
>> revisions.
> 
> I agree here.  The driver is advertising what IP blocks it can handle,
> so it makes sense to add both strings since it can handle both.

Wolfram,
so beside remarks about the compatibility strings. I didn't any other
comment since the v3 which was 7 weeks ago.

Does it mean that once I will have added the handle of this string,
you will be able to take the series for the 3.12 kernel?

Thanks,

> 
> thx,
> 
> Jason.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
Wolfram Sang Aug. 9, 2013, 1:22 p.m. UTC | #8
> Does it mean that once I will have added the handle of this string,
> you will be able to take the series for the 3.12 kernel?

Most looks good. The only thing I noticed from a glimpse is that maybe
you could use be32_to_cpu or something instead of shifting 8 bit chunks
out of the data_hi/lo?

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/i2c/busses/i2c-mv64xxx.c b/drivers/i2c/busses/i2c-mv64xxx.c
index b1f42bf..355ab64 100644
--- a/drivers/i2c/busses/i2c-mv64xxx.c
+++ b/drivers/i2c/busses/i2c-mv64xxx.c
@@ -55,6 +55,32 @@ 
 #define	MV64XXX_I2C_STATUS_MAST_RD_ADDR_2_NO_ACK	0xe8
 #define	MV64XXX_I2C_STATUS_NO_STATUS			0xf8
 
+/* Register defines (I2C bridge) */
+#define	MV64XXX_I2C_REG_TX_DATA_LO			0xc0
+#define	MV64XXX_I2C_REG_TX_DATA_HI			0xc4
+#define	MV64XXX_I2C_REG_RX_DATA_LO			0xc8
+#define	MV64XXX_I2C_REG_RX_DATA_HI			0xcc
+#define	MV64XXX_I2C_REG_BRIDGE_CONTROL			0xd0
+#define	MV64XXX_I2C_REG_BRIDGE_STATUS			0xd4
+#define	MV64XXX_I2C_REG_BRIDGE_INTR_CAUSE		0xd8
+#define	MV64XXX_I2C_REG_BRIDGE_INTR_MASK		0xdC
+#define	MV64XXX_I2C_REG_BRIDGE_TIMING			0xe0
+
+/* Bridge Control values */
+#define	MV64XXX_I2C_BRIDGE_CONTROL_WR			0x00000001
+#define	MV64XXX_I2C_BRIDGE_CONTROL_RD			0x00000002
+#define	MV64XXX_I2C_BRIDGE_CONTROL_ADDR_SHIFT		2
+#define	MV64XXX_I2C_BRIDGE_CONTROL_ADDR_EXT		0x00001000
+#define	MV64XXX_I2C_BRIDGE_CONTROL_TX_SIZE_SHIFT	13
+#define	MV64XXX_I2C_BRIDGE_CONTROL_RX_SIZE_SHIFT	16
+#define	MV64XXX_I2C_BRIDGE_CONTROL_ENABLE		0x00080000
+
+/* Bridge Status values */
+#define	MV64XXX_I2C_BRIDGE_STATUS_ERROR			0x00000001
+#define	MV64XXX_I2C_STATUS_OFFLOAD_ERROR		0xf0000001
+#define	MV64XXX_I2C_STATUS_OFFLOAD_OK			0xf0000000
+
+
 /* Driver states */
 enum {
 	MV64XXX_I2C_STATE_INVALID,
@@ -71,14 +97,17 @@  enum {
 enum {
 	MV64XXX_I2C_ACTION_INVALID,
 	MV64XXX_I2C_ACTION_CONTINUE,
+	MV64XXX_I2C_ACTION_OFFLOAD_SEND_START,
 	MV64XXX_I2C_ACTION_SEND_START,
 	MV64XXX_I2C_ACTION_SEND_RESTART,
+	MV64XXX_I2C_ACTION_OFFLOAD_RESTART,
 	MV64XXX_I2C_ACTION_SEND_ADDR_1,
 	MV64XXX_I2C_ACTION_SEND_ADDR_2,
 	MV64XXX_I2C_ACTION_SEND_DATA,
 	MV64XXX_I2C_ACTION_RCV_DATA,
 	MV64XXX_I2C_ACTION_RCV_DATA_STOP,
 	MV64XXX_I2C_ACTION_SEND_STOP,
+	MV64XXX_I2C_ACTION_OFFLOAD_SEND_STOP,
 };
 
 struct mv64xxx_i2c_regs {
@@ -117,6 +146,7 @@  struct mv64xxx_i2c_data {
 	spinlock_t		lock;
 	struct i2c_msg		*msg;
 	struct i2c_adapter	adapter;
+	bool			offload_enabled;
 };
 
 static struct mv64xxx_i2c_regs mv64xxx_i2c_regs_mv64xxx = {
@@ -165,6 +195,79 @@  mv64xxx_i2c_prepare_for_io(struct mv64xxx_i2c_data *drv_data,
 	}
 }
 
+static int mv64xxx_i2c_offload_msg(struct mv64xxx_i2c_data *drv_data)
+{
+	unsigned long data_reg_hi = 0;
+	unsigned long data_reg_lo = 0;
+	unsigned long ctrl_reg;
+	unsigned int i;
+	struct i2c_msg *msg = drv_data->msgs;
+
+	drv_data->msg = msg;
+	drv_data->byte_posn = 0;
+	drv_data->bytes_left = msg->len;
+	drv_data->aborting = 0;
+	drv_data->rc = 0;
+	/* Only regular transactions can be offloaded */
+	if ((msg->flags & ~(I2C_M_TEN | I2C_M_RD)) != 0)
+		return 1;
+
+	/* Only 1-8 byte transfers can be offloaded */
+	if (msg->len < 1 || msg->len > 8)
+		return 1;
+
+	/* Build transaction */
+	ctrl_reg = MV64XXX_I2C_BRIDGE_CONTROL_ENABLE |
+		   (msg->addr << MV64XXX_I2C_BRIDGE_CONTROL_ADDR_SHIFT);
+
+	if ((msg->flags & I2C_M_TEN) != 0)
+		ctrl_reg |=  MV64XXX_I2C_BRIDGE_CONTROL_ADDR_EXT;
+
+	if ((msg->flags & I2C_M_RD) == 0) {
+		for (i = 0; i < 4 && i < msg->len; i++)
+			data_reg_lo = data_reg_lo |
+					(msg->buf[i] << ((i & 0x3) * 8));
+
+		for (i = 4; i < 8 && i < msg->len; i++)
+			data_reg_hi = data_reg_hi |
+					(msg->buf[i] << ((i & 0x3) * 8));
+
+		ctrl_reg |= MV64XXX_I2C_BRIDGE_CONTROL_WR |
+		    (msg->len - 1) << MV64XXX_I2C_BRIDGE_CONTROL_TX_SIZE_SHIFT;
+	} else {
+		ctrl_reg |= MV64XXX_I2C_BRIDGE_CONTROL_RD |
+		    (msg->len - 1) << MV64XXX_I2C_BRIDGE_CONTROL_RX_SIZE_SHIFT;
+	}
+
+	/* Execute transaction */
+	writel_relaxed(data_reg_lo,
+		drv_data->reg_base + MV64XXX_I2C_REG_TX_DATA_LO);
+	writel_relaxed(data_reg_hi,
+		drv_data->reg_base + MV64XXX_I2C_REG_TX_DATA_HI);
+	writel(ctrl_reg, drv_data->reg_base + MV64XXX_I2C_REG_BRIDGE_CONTROL);
+
+	return 0;
+}
+
+static void
+mv64xxx_i2c_update_offload_data(struct i2c_msg *msg, unsigned long data_reg_hi,
+	unsigned long data_reg_lo)
+{
+	int i;
+
+	if ((msg->flags & I2C_M_RD) != 0) {
+		for (i = 0; i < 4 && i < msg->len; i++) {
+			msg->buf[i] = data_reg_lo & 0xFF;
+			data_reg_lo >>= 8;
+		}
+
+		for (i = 4; i < 8 && i < msg->len; i++) {
+			msg->buf[i] = data_reg_hi & 0xFF;
+			data_reg_hi >>= 8;
+		}
+	}
+
+}
 /*
  *****************************************************************************
  *
@@ -177,6 +280,15 @@  mv64xxx_i2c_prepare_for_io(struct mv64xxx_i2c_data *drv_data,
 static void
 mv64xxx_i2c_hw_init(struct mv64xxx_i2c_data *drv_data)
 {
+	if (drv_data->offload_enabled) {
+		writel(0, drv_data->reg_base + MV64XXX_I2C_REG_BRIDGE_CONTROL);
+		writel(0, drv_data->reg_base + MV64XXX_I2C_REG_BRIDGE_TIMING);
+		writel(0, drv_data->reg_base +
+			MV64XXX_I2C_REG_BRIDGE_INTR_CAUSE);
+		writel(0, drv_data->reg_base +
+			MV64XXX_I2C_REG_BRIDGE_INTR_MASK);
+	}
+
 	writel(0, drv_data->reg_base + drv_data->reg_offsets.soft_reset);
 	writel(MV64XXX_I2C_BAUD_DIV_M(drv_data->freq_m) | MV64XXX_I2C_BAUD_DIV_N(drv_data->freq_n),
 		drv_data->reg_base + drv_data->reg_offsets.clock);
@@ -283,6 +395,16 @@  mv64xxx_i2c_fsm(struct mv64xxx_i2c_data *drv_data, u32 status)
 		drv_data->rc = -ENXIO;
 		break;
 
+	case MV64XXX_I2C_STATUS_OFFLOAD_OK:
+		if (drv_data->send_stop || drv_data->aborting) {
+			drv_data->action = MV64XXX_I2C_ACTION_OFFLOAD_SEND_STOP;
+			drv_data->state = MV64XXX_I2C_STATE_IDLE;
+		} else {
+			drv_data->action = MV64XXX_I2C_ACTION_OFFLOAD_RESTART;
+			drv_data->state = MV64XXX_I2C_STATE_WAITING_FOR_RESTART;
+		}
+		break;
+
 	default:
 		dev_err(&drv_data->adapter.dev,
 			"mv64xxx_i2c_fsm: Ctlr Error -- state: 0x%x, "
@@ -298,21 +420,36 @@  mv64xxx_i2c_fsm(struct mv64xxx_i2c_data *drv_data, u32 status)
 static void
 mv64xxx_i2c_do_action(struct mv64xxx_i2c_data *drv_data)
 {
+	unsigned long data_reg_hi = 0;
+	unsigned long data_reg_lo = 0;
+
 	switch(drv_data->action) {
+	case MV64XXX_I2C_ACTION_OFFLOAD_RESTART:
+		data_reg_lo = readl(drv_data->reg_base +
+				MV64XXX_I2C_REG_RX_DATA_LO);
+		data_reg_hi = readl(drv_data->reg_base +
+				MV64XXX_I2C_REG_RX_DATA_HI);
+		mv64xxx_i2c_update_offload_data(drv_data->msg, data_reg_hi,
+						data_reg_lo);
+		writel(0, drv_data->reg_base +	MV64XXX_I2C_REG_BRIDGE_CONTROL);
+		writel(0, drv_data->reg_base +
+			MV64XXX_I2C_REG_BRIDGE_INTR_CAUSE);
+		/* FALLTHRU */
 	case MV64XXX_I2C_ACTION_SEND_RESTART:
 		/* We should only get here if we have further messages */
 		BUG_ON(drv_data->num_msgs == 0);
 
-		drv_data->cntl_bits |= MV64XXX_I2C_REG_CONTROL_START;
-		writel(drv_data->cntl_bits,
-			drv_data->reg_base + drv_data->reg_offsets.control);
-
 		drv_data->msgs++;
 		drv_data->num_msgs--;
+		if (!(drv_data->offload_enabled &&
+				mv64xxx_i2c_offload_msg(drv_data))) {
+			drv_data->cntl_bits |= MV64XXX_I2C_REG_CONTROL_START;
+			writel(drv_data->cntl_bits,
+			drv_data->reg_base + drv_data->reg_offsets.control);
 
-		/* Setup for the next message */
-		mv64xxx_i2c_prepare_for_io(drv_data, drv_data->msgs);
-
+			/* Setup for the next message */
+			mv64xxx_i2c_prepare_for_io(drv_data, drv_data->msgs);
+		}
 		/*
 		 * We're never at the start of the message here, and by this
 		 * time it's already too late to do any protocol mangling.
@@ -326,6 +463,12 @@  mv64xxx_i2c_do_action(struct mv64xxx_i2c_data *drv_data)
 			drv_data->reg_base + drv_data->reg_offsets.control);
 		break;
 
+	case MV64XXX_I2C_ACTION_OFFLOAD_SEND_START:
+		if (mv64xxx_i2c_offload_msg(drv_data) <= 0)
+			break;
+		else
+			drv_data->action = MV64XXX_I2C_ACTION_SEND_START;
+		/* FALLTHRU */
 	case MV64XXX_I2C_ACTION_SEND_START:
 		writel(drv_data->cntl_bits | MV64XXX_I2C_REG_CONTROL_START,
 			drv_data->reg_base + drv_data->reg_offsets.control);
@@ -375,6 +518,7 @@  mv64xxx_i2c_do_action(struct mv64xxx_i2c_data *drv_data)
 			"mv64xxx_i2c_do_action: Invalid action: %d\n",
 			drv_data->action);
 		drv_data->rc = -EIO;
+
 		/* FALLTHRU */
 	case MV64XXX_I2C_ACTION_SEND_STOP:
 		drv_data->cntl_bits &= ~MV64XXX_I2C_REG_CONTROL_INTEN;
@@ -383,6 +527,20 @@  mv64xxx_i2c_do_action(struct mv64xxx_i2c_data *drv_data)
 		drv_data->block = 0;
 		wake_up(&drv_data->waitq);
 		break;
+
+	case MV64XXX_I2C_ACTION_OFFLOAD_SEND_STOP:
+		data_reg_lo = readl(drv_data->reg_base +
+				MV64XXX_I2C_REG_RX_DATA_LO);
+		data_reg_hi = readl(drv_data->reg_base +
+				MV64XXX_I2C_REG_RX_DATA_HI);
+		mv64xxx_i2c_update_offload_data(drv_data->msg, data_reg_hi,
+						data_reg_lo);
+		writel(0, drv_data->reg_base +	MV64XXX_I2C_REG_BRIDGE_CONTROL);
+		writel(0, drv_data->reg_base +
+			MV64XXX_I2C_REG_BRIDGE_INTR_CAUSE);
+		drv_data->block = 0;
+		wake_up(&drv_data->waitq);
+		break;
 	}
 }
 
@@ -395,6 +553,21 @@  mv64xxx_i2c_intr(int irq, void *dev_id)
 	irqreturn_t	rc = IRQ_NONE;
 
 	spin_lock_irqsave(&drv_data->lock, flags);
+
+	if (drv_data->offload_enabled) {
+		while (readl(drv_data->reg_base +
+				MV64XXX_I2C_REG_BRIDGE_INTR_CAUSE)) {
+			int reg_status = readl(drv_data->reg_base +
+					MV64XXX_I2C_REG_BRIDGE_STATUS);
+			if (reg_status & MV64XXX_I2C_BRIDGE_STATUS_ERROR)
+				status = MV64XXX_I2C_STATUS_OFFLOAD_ERROR;
+			else
+				status = MV64XXX_I2C_STATUS_OFFLOAD_OK;
+			mv64xxx_i2c_fsm(drv_data, status);
+			mv64xxx_i2c_do_action(drv_data);
+			rc = IRQ_HANDLED;
+		}
+	}
 	while (readl(drv_data->reg_base + drv_data->reg_offsets.control) &
 						MV64XXX_I2C_REG_CONTROL_IFLG) {
 		status = readl(drv_data->reg_base + drv_data->reg_offsets.status);
@@ -459,11 +632,15 @@  mv64xxx_i2c_execute_msg(struct mv64xxx_i2c_data *drv_data, struct i2c_msg *msg,
 	unsigned long	flags;
 
 	spin_lock_irqsave(&drv_data->lock, flags);
-	mv64xxx_i2c_prepare_for_io(drv_data, msg);
-
-	drv_data->action = MV64XXX_I2C_ACTION_SEND_START;
-	drv_data->state = MV64XXX_I2C_STATE_WAITING_FOR_START_COND;
+	if (drv_data->offload_enabled) {
+		drv_data->action = MV64XXX_I2C_ACTION_OFFLOAD_SEND_START;
+		drv_data->state = MV64XXX_I2C_STATE_WAITING_FOR_START_COND;
+	} else {
+		mv64xxx_i2c_prepare_for_io(drv_data, msg);
 
+		drv_data->action = MV64XXX_I2C_ACTION_SEND_START;
+		drv_data->state = MV64XXX_I2C_STATE_WAITING_FOR_START_COND;
+	}
 	drv_data->send_stop = is_last;
 	drv_data->block = 1;
 	mv64xxx_i2c_do_action(drv_data);
@@ -601,6 +778,13 @@  mv64xxx_of_config(struct mv64xxx_i2c_data *drv_data,
 
 	memcpy(&drv_data->reg_offsets, device->data, sizeof(drv_data->reg_offsets));
 
+	/*
+	 * For controllers embedded in new SoCs activate the
+	 * Transaction Generator support.
+	 */
+	if (of_device_is_compatible(np, "marvell,mv78230-i2c"))
+		drv_data->offload_enabled = true;
+
 out:
 	return rc;
 #endif
@@ -654,6 +838,7 @@  mv64xxx_i2c_probe(struct platform_device *pd)
 		drv_data->freq_n = pdata->freq_n;
 		drv_data->irq = platform_get_irq(pd, 0);
 		drv_data->adapter.timeout = msecs_to_jiffies(pdata->timeout);
+		drv_data->offload_enabled = 0;
 		memcpy(&drv_data->reg_offsets, &mv64xxx_i2c_regs_mv64xxx, sizeof(drv_data->reg_offsets));
 	} else if (pd->dev.of_node) {
 		rc = mv64xxx_of_config(drv_data, &pd->dev);