Skip to contentSkip to conversation

i2c: tegra: Share same DMA channel for Rx and Tx

PATCH Superseded Akhil R 2 participants 3 messages Feb 22, 2023 – Feb 23, 2023

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Akhil R

author
#0 lore raw

PATCH v2 1/1 i2c: tegra: Share same DMA channel for Rx and Tx

Allocate only one DMA channel for I2C and share it for both Tx and Rx. Since I2C supports only half duplex, there is no impact on perf with this.

---
v1->v2: Remove WARN_ON for DMA channel mismatch. There is only one channel in use with this change.

 drivers/i2c/busses/i2c-tegra.c | 54 ++++++++++------------------------
 1 file changed, 15 insertions(+), 39 deletions(-)
drivers/i2c/busses/i2c-tegra.c +15 -39
index 6aab84c8d22b..f52b835f1700 100644
@@ -248,8 +248,7 @@ struct tegra_i2c_hw_feature {
  * @msg_read: indicates that the transfer is a read access
  * @timings: i2c timings information like bus frequency
  * @multimaster_mode: indicates that I2C controller is in multi-master mode
- * @tx_dma_chan: DMA transmit channel
- * @rx_dma_chan: DMA receive channel
+ * @dma_chan: DMA channel
  * @dma_phys: handle to DMA resources
  * @dma_buf: pointer to allocated DMA buffer
  * @dma_buf_size: DMA buffer size
@@ -281,8 +280,7 @@ struct tegra_i2c_dev {
 	u8 *msg_buf;
 
 	struct completion dma_complete;
-	struct dma_chan *tx_dma_chan;
-	struct dma_chan *rx_dma_chan;
+	struct dma_chan *dma_chan;
 	unsigned int dma_buf_size;
 	struct device *dma_dev;
 	dma_addr_t dma_phys;
@@ -398,7 +396,7 @@ static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
 	reinit_completion(&i2c_dev->dma_complete);
 
 	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
-	chan = i2c_dev->msg_read ? i2c_dev->rx_dma_chan : i2c_dev->tx_dma_chan;
+	chan = i2c_dev->dma_chan;
 
 	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
 					       len, dir, DMA_PREP_INTERRUPT |
@@ -426,14 +424,9 @@ static void tegra_i2c_release_dma(struct tegra_i2c_dev *i2c_dev)
 		i2c_dev->dma_buf = NULL;
 	}
 
-	if (i2c_dev->tx_dma_chan) {
-		dma_release_channel(i2c_dev->tx_dma_chan);
-		i2c_dev->tx_dma_chan = NULL;
-	}
-
-	if (i2c_dev->rx_dma_chan) {
-		dma_release_channel(i2c_dev->rx_dma_chan);
-		i2c_dev->rx_dma_chan = NULL;
+	if (i2c_dev->dma_chan) {
+		dma_release_channel(i2c_dev->dma_chan);
+		i2c_dev->dma_chan = NULL;
 	}
 }
 
@@ -457,25 +450,18 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
 		return 0;
 	}
 
-	chan = dma_request_chan(i2c_dev->dev, "rx");
-	if (IS_ERR(chan)) {
-		err = PTR_ERR(chan);
-		goto err_out;
-	}
-
-	i2c_dev->rx_dma_chan = chan;
-
+	/* The same channel will be used for both Rx and Tx.
+	 * Keeping the name as tx for backward compatibility with
+	 * existing devicetrees.
+	 */
 	chan = dma_request_chan(i2c_dev->dev, "tx");
 	if (IS_ERR(chan)) {
 		err = PTR_ERR(chan);
 		goto err_out;
 	}
 
-	i2c_dev->tx_dma_chan = chan;
-
-	WARN_ON(i2c_dev->tx_dma_chan->device != i2c_dev->rx_dma_chan->device);
+	i2c_dev->dma_chan = chan;
 	i2c_dev->dma_dev = chan->device->dev;
-
 	i2c_dev->dma_buf_size = i2c_dev->hw->quirks->max_write_len +
 				I2C_PACKET_HEADER_SIZE;
 
@@ -974,11 +960,7 @@ static irqreturn_t tegra_i2c_isr(int irq, void *dev_id)
 		dvc_writel(i2c_dev, DVC_STATUS_I2C_DONE_INTR, DVC_STATUS);
 
 	if (i2c_dev->dma_mode) {
-		if (i2c_dev->msg_read)
-			dmaengine_terminate_async(i2c_dev->rx_dma_chan);
-		else
-			dmaengine_terminate_async(i2c_dev->tx_dma_chan);
-
+		dmaengine_terminate_async(i2c_dev->dma_chan);
 		complete(&i2c_dev->dma_complete);
 	}
 
@@ -1008,8 +990,8 @@ static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
 		else
 			dma_burst = 8;
 
+		chan = i2c_dev->dma_chan;
 		if (i2c_dev->msg_read) {
-			chan = i2c_dev->rx_dma_chan;
 			reg_offset = tegra_i2c_reg_addr(i2c_dev, I2C_RX_FIFO);
 
 			slv_config.src_addr = i2c_dev->base_phys + reg_offset;
@@ -1021,7 +1003,6 @@ static void tegra_i2c_config_fifo_trig(struct tegra_i2c_dev *i2c_dev,
 			else
 				val = I2C_FIFO_CONTROL_RX_TRIG(dma_burst);
 		} else {
-			chan = i2c_dev->tx_dma_chan;
 			reg_offset = tegra_i2c_reg_addr(i2c_dev, I2C_TX_FIFO);
 
 			slv_config.dst_addr = i2c_dev->base_phys + reg_offset;
@@ -1333,13 +1314,8 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
 		 * performs synchronization after the transfer's termination
 		 * and we want to get a completion if transfer succeeded.
 		 */
-		dmaengine_synchronize(i2c_dev->msg_read ?
-				      i2c_dev->rx_dma_chan :
-				      i2c_dev->tx_dma_chan);
-
-		dmaengine_terminate_sync(i2c_dev->msg_read ?
-					 i2c_dev->rx_dma_chan :
-					 i2c_dev->tx_dma_chan);
+		dmaengine_synchronize(i2c_dev->dma_chan);
+		dmaengine_terminate_sync(i2c_dev->dma_chan);
 
 		if (!time_left && !completion_done(&i2c_dev->dma_complete)) {
 			dev_err(i2c_dev->dev, "DMA transfer timed out\n");
Signature · 1 line
2.17.1
#1 lore raw
Quoted 11 lines · Akhil R

Allocate only one DMA channel for I2C and share it for both Tx and Rx. Since I2C supports only half duplex, there is no impact on perf with this.

---
v1->v2: Remove WARN_ON for DMA channel mismatch. There is only one channel in use with this change.

 drivers/i2c/busses/i2c-tegra.c | 54 ++++++++++------------------------
 1 file changed, 15 insertions(+), 39 deletions(-)

I'm a little confused by this. All device trees already list the very same reference for both TX and RX DMA channels in the I2C nodes, so these channels are already effectively shared, aren't they?

So all this does is to get rid of the duplicated pointer? In practice, is the DMA channel pointer going to point to the exact same memory or are these separate objects that happen to point to the same hardware resource?

In either case, I think the commit message should clarify that. Also, a few minor nits below...

Quoted 25 lines
drivers/i2c/busses/i2c-tegra.c +2 -4
index 6aab84c8d22b..f52b835f1700 100644
@@ -248,8 +248,7 @@ struct tegra_i2c_hw_feature {
  * @msg_read: indicates that the transfer is a read access
  * @timings: i2c timings information like bus frequency
  * @multimaster_mode: indicates that I2C controller is in multi-master mode
- * @tx_dma_chan: DMA transmit channel
- * @rx_dma_chan: DMA receive channel
+ * @dma_chan: DMA channel
  * @dma_phys: handle to DMA resources
  * @dma_buf: pointer to allocated DMA buffer
  * @dma_buf_size: DMA buffer size
@@ -281,8 +280,7 @@ struct tegra_i2c_dev {
 	u8 *msg_buf;
 
 	struct completion dma_complete;
-	struct dma_chan *tx_dma_chan;
-	struct dma_chan *rx_dma_chan;
+	struct dma_chan *dma_chan;
 	unsigned int dma_buf_size;
 	struct device *dma_dev;
 	dma_addr_t dma_phys;

drivers/i2c/busses/i2c-tegra.c static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len) · line 399

Quoted 5 lines
drivers/i2c/busses/i2c-tegra.c +1 -1
@@ -398,7 +396,7 @@ static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
 	reinit_completion(&i2c_dev->dma_complete);
 
 	dir = i2c_dev->msg_read ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV;
-	chan = i2c_dev->msg_read ? i2c_dev->rx_dma_chan : i2c_dev->tx_dma_chan;
+	chan = i2c_dev->dma_chan;

Is there any point in keeping around the local variable? It doesn't serve any purpose anymore.

Quoted 27 lines
drivers/i2c/busses/i2c-tegra.c +3 -10
 	dma_desc = dmaengine_prep_slave_single(chan, i2c_dev->dma_phys,
 					       len, dir, DMA_PREP_INTERRUPT |
@@ -426,14 +424,9 @@ static void tegra_i2c_release_dma(struct tegra_i2c_dev *i2c_dev)
 		i2c_dev->dma_buf = NULL;
 	}
 
-	if (i2c_dev->tx_dma_chan) {
-		dma_release_channel(i2c_dev->tx_dma_chan);
-		i2c_dev->tx_dma_chan = NULL;
-	}
-
-	if (i2c_dev->rx_dma_chan) {
-		dma_release_channel(i2c_dev->rx_dma_chan);
-		i2c_dev->rx_dma_chan = NULL;
+	if (i2c_dev->dma_chan) {
+		dma_release_channel(i2c_dev->dma_chan);
+		i2c_dev->dma_chan = NULL;
 	}
 }
 
@@ -457,25 +450,18 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
 		return 0;
 	}
 
-	chan = dma_request_chan(i2c_dev->dev, "rx");
-	if (IS_ERR(chan)) {

drivers/i2c/busses/i2c-tegra.c static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev) · line 456

Quoted 10 lines
drivers/i2c/busses/i2c-tegra.c +4 -6
-		err = PTR_ERR(chan);
-		goto err_out;
-	}
-
-	i2c_dev->rx_dma_chan = chan;
-
+	/* The same channel will be used for both Rx and Tx.
+	 * Keeping the name as tx for backward compatibility with
+	 * existing devicetrees.
+	 */

Block comments should have the starting /* on a line of its own. Also, s/Rx/RX/ and s/Tx/TX/ for consistency. Looks like you also used "Rx" and "Tx" in the commit message. The driver uses "RX" and "TX" elsewhere, so better switch to that in both this comment and the commit message.

Also, perhaps quote the "tx" reference above to highlight that it's the literal string in DT that you're referring to.

Thierry

author
#2 lore raw
Quoted 9 lines
Quoted 9 lines

Allocate only one DMA channel for I2C and share it for both Tx and Rx. Since I2C supports only half duplex, there is no impact on perf with this.

---
v1->v2: Remove WARN_ON for DMA channel mismatch. There is only one channel in use with this change.

Quoted 6 lines
Quoted 2 lines
 drivers/i2c/busses/i2c-tegra.c | 54 ++++++++++------------------------
 1 file changed, 15 insertions(+), 39 deletions(-)

I'm a little confused by this. All device trees already list the very same reference for both TX and RX DMA channels in the I2C nodes, so these channels are already effectively shared, aren't they?

The value given in DT refers to the slave-id of an I2C instance and not the channel number. Each I2C instance was in-effect using two different DMA channels with the same slave-id.
This change is to free up the extra DMA channel since both will not be used simultaneously.

Quoted 5 lines

So all this does is to get rid of the duplicated pointer? In practice, is the DMA channel pointer going to point to the exact same memory or are these separate objects that happen to point to the same hardware resource?

They are indeed separate hardware resource itself. As I described above, two separate DMA channels were in use for each I2C instance.

Quoted 3 lines

In either case, I think the commit message should clarify that. Also, a few minor nits below...

Agreed with the other comments.

Regards,
Akhil