Skip to contentSkip to conversation

i2c: designware: Do not complete i2c read without RX_FULL interrupt

PATCH Superseded [email protected] 2 participants 4 messages Dec 14, 2021

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 i2c: designware: Do not complete i2c read without RX_FULL interrupt

From: Tamal Saha <[email protected]>

Intel Keem Bay platform supports multimaster operations over same i2c bus using Synopsys i2c DesignWare IP. When multi masters initiate i2c operation simultaneously in a loop, SCL line is stucked low forever after few i2c operations. Following interrupt sequences are observed in:
working case: TX_EMPTY, RX_FULL and STOP_DET
non working case: TX_EMPTY, STOP_DET, RX_FULL.

DW_apb_i2c stretches the SCL line when the TX FIFO is empty or when RX FIFO is full. The DW_apb_i2c master will continue to hold the SCL line LOW until RX FIFO is read.

Linux kernel i2c DesignWare driver does not handle above non working sequence. TX_EMPTY, RX_FULL and STOP_DET routine execution are required in sequence although RX_FULL interrupt is raised after STOP_DET by hardware. Clear STOP_DET for the following conditions:
(STOP_DET ,RX_FULL, rx_outstanding)

    Write Operation: (1, 0, 0)
    Read Operation:
      RX_FULL followed by STOP_DET: (0, 1, 1) -> (1, 0, 0)
      STOP_DET followed by RX_FULL: (1, 0, 1) -> (1, 1, 0)
      RX_FULL and STOP_DET together: (1, 1, 1)
---
 drivers/i2c/busses/i2c-designware-master.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/i2c/busses/i2c-designware-master.c +5 -2
index 9b08bb5df38d..9177463c2cbb 100644
@@ -701,7 +701,8 @@ static u32 i2c_dw_read_clear_intrbits(struct dw_i2c_dev *dev)
 		regmap_read(dev->map, DW_IC_CLR_RX_DONE, &dummy);
 	if (stat & DW_IC_INTR_ACTIVITY)
 		regmap_read(dev->map, DW_IC_CLR_ACTIVITY, &dummy);
-	if (stat & DW_IC_INTR_STOP_DET)
+	if ((stat & DW_IC_INTR_STOP_DET) &&
+	    ((dev->rx_outstanding == 0) || (stat & DW_IC_INTR_RX_FULL)))
 		regmap_read(dev->map, DW_IC_CLR_STOP_DET, &dummy);
 	if (stat & DW_IC_INTR_START_DET)
 		regmap_read(dev->map, DW_IC_CLR_START_DET, &dummy);
@@ -723,6 +724,7 @@ static int i2c_dw_irq_handler_master(struct dw_i2c_dev *dev)
 	if (stat & DW_IC_INTR_TX_ABRT) {
 		dev->cmd_err |= DW_IC_ERR_TX_ABRT;
 		dev->status = STATUS_IDLE;
+		dev->rx_outstanding = 0;
 
 		/*
 		 * Anytime TX_ABRT is set, the contents of the tx/rx
@@ -745,7 +747,8 @@ static int i2c_dw_irq_handler_master(struct dw_i2c_dev *dev)
 	 */
 
 tx_aborted:
-	if ((stat & (DW_IC_INTR_TX_ABRT | DW_IC_INTR_STOP_DET)) || dev->msg_err)
+	if (((stat & (DW_IC_INTR_TX_ABRT | DW_IC_INTR_STOP_DET)) || dev->msg_err) &&
+	     (dev->rx_outstanding == 0))
 		complete(&dev->cmd_complete);
 	else if (unlikely(dev->flags & ACCESS_INTR_MASK)) {
 		/* Workaround to trigger pending interrupt */
Signature · 1 line
2.17.1
#1 lore raw
Quoted 19 lines · On Tue, Dec 14, 2021 at 08:25:18AM +0530, [email protected] wrote:

From: Tamal Saha <[email protected]>

Intel Keem Bay platform supports multimaster operations over same i2c bus using Synopsys i2c DesignWare IP. When multi masters initiate i2c operation simultaneously in a loop, SCL line is stucked low forever after few i2c operations. Following interrupt sequences are observed in:
working case: TX_EMPTY, RX_FULL and STOP_DET
non working case: TX_EMPTY, STOP_DET, RX_FULL.

DW_apb_i2c stretches the SCL line when the TX FIFO is empty or when RX FIFO is full. The DW_apb_i2c master will continue to hold the SCL line LOW until RX FIFO is read.

Linux kernel i2c DesignWare driver does not handle above non working sequence. TX_EMPTY, RX_FULL and STOP_DET routine execution are required in sequence although RX_FULL interrupt is raised after STOP_DET by hardware. Clear STOP_DET for the following conditions:
(STOP_DET ,RX_FULL, rx_outstanding)

Quoted 8 lines
    Write Operation: (1, 0, 0)
    Read Operation:
      RX_FULL followed by STOP_DET: (0, 1, 1) -> (1, 0, 0)
      STOP_DET followed by RX_FULL: (1, 0, 1) -> (1, 1, 0)
      RX_FULL and STOP_DET together: (1, 1, 1)

---

So, where is the changelog?

Quoted 38 lines
 drivers/i2c/busses/i2c-designware-master.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/i2c/busses/i2c-designware-master.c +5 -2
index 9b08bb5df38d..9177463c2cbb 100644
@@ -701,7 +701,8 @@ static u32 i2c_dw_read_clear_intrbits(struct dw_i2c_dev *dev)
 		regmap_read(dev->map, DW_IC_CLR_RX_DONE, &dummy);
 	if (stat & DW_IC_INTR_ACTIVITY)
 		regmap_read(dev->map, DW_IC_CLR_ACTIVITY, &dummy);
-	if (stat & DW_IC_INTR_STOP_DET)
+	if ((stat & DW_IC_INTR_STOP_DET) &&
+	    ((dev->rx_outstanding == 0) || (stat & DW_IC_INTR_RX_FULL)))
 		regmap_read(dev->map, DW_IC_CLR_STOP_DET, &dummy);
 	if (stat & DW_IC_INTR_START_DET)
 		regmap_read(dev->map, DW_IC_CLR_START_DET, &dummy);
@@ -723,6 +724,7 @@ static int i2c_dw_irq_handler_master(struct dw_i2c_dev *dev)
 	if (stat & DW_IC_INTR_TX_ABRT) {
 		dev->cmd_err |= DW_IC_ERR_TX_ABRT;
 		dev->status = STATUS_IDLE;
+		dev->rx_outstanding = 0;
 
 		/*
 		 * Anytime TX_ABRT is set, the contents of the tx/rx
@@ -745,7 +747,8 @@ static int i2c_dw_irq_handler_master(struct dw_i2c_dev *dev)
 	 */
 
 tx_aborted:
-	if ((stat & (DW_IC_INTR_TX_ABRT | DW_IC_INTR_STOP_DET)) || dev->msg_err)
+	if (((stat & (DW_IC_INTR_TX_ABRT | DW_IC_INTR_STOP_DET)) || dev->msg_err) &&
+	     (dev->rx_outstanding == 0))
 		complete(&dev->cmd_complete);
 	else if (unlikely(dev->flags & ACCESS_INTR_MASK)) {
 		/* Workaround to trigger pending interrupt */
Signature · 1 line
2.17.1
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#2 lore raw
Quoted 4 lines · Andy Shevchenko
Quoted 3 lines · On Tue, Dec 14, 2021 at 08:25:18AM +0530, [email protected] wrote:

From: Tamal Saha <[email protected]>

Intel Keem Bay platform supports multimaster operations over same i2c

Here you put multimaster...

Quoted 1 line
Quoted 1 line

bus using Synopsys i2c DesignWare IP. When multi masters initiate i2c

...and here multi masters.

Be consistent.

I believe the best way is to use dash variant, i.e. multi-master(s),

Quoted 25 lines
Quoted 15 lines

operation simultaneously in a loop, SCL line is stucked low forever after few i2c operations. Following interrupt sequences are observed in:
working case: TX_EMPTY, RX_FULL and STOP_DET
non working case: TX_EMPTY, STOP_DET, RX_FULL.

DW_apb_i2c stretches the SCL line when the TX FIFO is empty or when RX FIFO is full. The DW_apb_i2c master will continue to hold the SCL line LOW until RX FIFO is read.

Linux kernel i2c DesignWare driver does not handle above non working sequence. TX_EMPTY, RX_FULL and STOP_DET routine execution are required in sequence although RX_FULL interrupt is raised after STOP_DET by hardware. Clear STOP_DET for the following conditions:
(STOP_DET ,RX_FULL, rx_outstanding)

Quoted 8 lines
    Write Operation: (1, 0, 0)
    Read Operation:
      RX_FULL followed by STOP_DET: (0, 1, 1) -> (1, 0, 0)
      STOP_DET followed by RX_FULL: (1, 0, 1) -> (1, 1, 0)
      RX_FULL and STOP_DET together: (1, 1, 1)

---

So, where is the changelog?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#3 lore raw

Thanks Andy. I'll incorporate patch version changelog and review findings for this patch in next version. Please let me know, if you have more findings for this patch.

Quoted 57 lines

-----Original Message-----
From: Andy Shevchenko <[email protected]>
Sent: Tuesday, 14 December, 2021 5:33 PM
To: Saha, Tamal <[email protected]>
Cc: [email protected]; [email protected]; linux- [email protected]; Senthil, Bala <[email protected]>
Subject: Re: [PATCH v2] i2c: designware: Do not complete i2c read without RX_FULL interrupt

Quoted 5 lines · Andy Shevchenko
Quoted 4 lines · On Tue, Dec 14, 2021 at 08:25:18AM +0530, [email protected] wrote:

From: Tamal Saha <[email protected]>

Intel Keem Bay platform supports multimaster operations over same i2c

Here you put multimaster...

Quoted 2 lines
Quoted 2 lines

bus using Synopsys i2c DesignWare IP. When multi masters initiate i2c

...and here multi masters.

Be consistent.

I believe the best way is to use dash variant, i.e. multi-master(s),

Quoted 21 lines
Quoted 21 lines

operation simultaneously in a loop, SCL line is stucked low forever after few i2c operations. Following interrupt sequences are observed in:
working case: TX_EMPTY, RX_FULL and STOP_DET
non working case: TX_EMPTY, STOP_DET, RX_FULL.

DW_apb_i2c stretches the SCL line when the TX FIFO is empty or when RX FIFO is full. The DW_apb_i2c master will continue to hold the SCL line LOW until RX FIFO is read.

Linux kernel i2c DesignWare driver does not handle above non working sequence. TX_EMPTY, RX_FULL and STOP_DET routine execution are required in sequence although RX_FULL interrupt is raised after STOP_DET by hardware. Clear STOP_DET for the following conditions:
(STOP_DET ,RX_FULL, rx_outstanding)

    Write Operation: (1, 0, 0)
    Read Operation:
      RX_FULL followed by STOP_DET: (0, 1, 1) -> (1, 0, 0)
      STOP_DET followed by RX_FULL: (1, 0, 1) -> (1, 1, 0)
      RX_FULL and STOP_DET together: (1, 1, 1)
Quoted 4 lines
Quoted 2 lines

---

So, where is the changelog?

--
With Best Regards,
Andy Shevchenko