Skip to contentSkip to conversation

i2c: aspeed: Add H/W timeout support

PATCH Superseded Jae Hyun Yoo 5 participants 23 messages Sep 4, 2019 – Sep 12, 2019

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover i2c: aspeed: Add H/W timeout support

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Jae Hyun Yoo (2):
dt-bindings: i2c: aspeed: add hardware timeout support
i2c: aspeed: add slave inactive timeout support

 .../devicetree/bindings/i2c/i2c-aspeed.txt    |  2 +
 drivers/i2c/busses/i2c-aspeed.c               | 79 +++++++++++++++++--
 2 files changed, 75 insertions(+), 6 deletions(-)
Signature · 1 line
2.23.0
#1 lore raw

PATCH 1/2 dt-bindings: i2c: aspeed: add hardware timeout support

Append a binding to support hardware timeout feature.

---
 Documentation/devicetree/bindings/i2c/i2c-aspeed.txt | 2 ++
 1 file changed, 2 insertions(+)
Documentation/devicetree/bindings/i2c/i2c-aspeed.txt +2 -0
index e5b46885c15e..71de956ffa4f 100644
@@ -29,6 +29,8 @@ Optional Properties:
 			  DMA mode instead of PIO or FIFO respectively, I2C
 			  can't use DMA mode. IF both DMA and buffer modes are
 			  enabled, DMA mode will be selected.
+- aspeed,hw-timeout-ms	: Hardware timeout in milliseconds. If it's not
+			  specified, the H/W timeout feature will be disabled.
 
 Example:
 
Signature · 1 line
2.23.0
#2 lore raw

PATCH 2/2 i2c: aspeed: add slave inactive timeout support

In reply to Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

---
 drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---
 1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +73 -6
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
 #define ASPEED_I2CD_TIME_SCL_HIGH_MASK			GENMASK(19, 16)
 #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT			12
 #define ASPEED_I2CD_TIME_SCL_LOW_MASK			GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT	8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK	GENMASK(9, 8)
 #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK		GENMASK(3, 0)
 #define ASPEED_I2CD_TIME_SCL_REG_MAX			GENMASK(3, 0)
+
 /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL				0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT		0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK			GENMASK(4, 0)
 
 /* 0x0c : I2CD Interrupt Control Register &
  * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
  * These share bit definitions, so use the same values for the enable &
  * status bits.
  */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT		BIT(15)
 #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT			BIT(14)
 #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE		BIT(13)
 #define ASPEED_I2CD_INTR_SLAVE_MATCH			BIT(7)
@@ -96,8 +101,11 @@
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
 		 ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS					       \
+		ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
 #define ASPEED_I2CD_INTR_ALL						       \
-		(ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
+		(ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |		       \
+		 ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
 		 ASPEED_I2CD_INTR_BUS_RECOVER_DONE |			       \
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
 							   u32 divisor);
 	unsigned long			parent_clk_frequency;
 	u32				bus_frequency;
+	u32				hw_timeout_ms;
 	/* Transaction state. */
 	enum aspeed_i2c_master_state	master_state;
 	struct i2c_msg			*msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
 }
 
 #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+	if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+		return -EIO;
+
+	return 0;
+}
+
 static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 {
 	u32 command, irq_handled = 0;
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	if (!slave)
 		return 0;
 
+	if (aspeed_i2c_check_slave_error(irq_status)) {
+		dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+			irq_status);
+		irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
+		return irq_handled;
+	}
+
 	command = readl(bus->base + ASPEED_I2C_CMD_REG);
 
 	/* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
 	}
 }
 
-static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
 {
 	if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
 		return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	 * should clear the command queue effectively taking us back to the
 	 * INACTIVE state.
 	 */
-	ret = aspeed_i2c_is_irq_error(irq_status);
+	ret = aspeed_i2c_check_master_error(irq_status);
 	if (ret) {
-		dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+		dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
 			irq_status);
 		irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
 		if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
 /* precondition: bus.lock has been acquired. */
 static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 {
+	u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
 	u32 divisor, clk_reg_val;
 
 	divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 			ASPEED_I2CD_TIME_THDSTA_MASK |
 			ASPEED_I2CD_TIME_TACST_MASK);
 	clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);
+
+	if (bus->hw_timeout_ms) {
+		u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+			     ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+		u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+				ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+		timeout_base_divisor = 0;
+
+		do {
+			timeout_tick_us = 1000 * (16384 <<
+						  (timeout_base_divisor << 1)) /
+					  (bus->parent_clk_frequency / 1000);
+
+			if (timeout_base_divisor == div_max ||
+			    timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+			    bus->hw_timeout_ms * 1000)
+				break;
+		} while (timeout_base_divisor++ < div_max);
+
+		if (timeout_tick_us) {
+			timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+						      timeout_tick_us);
+			if (timeout_cycles == 0)
+				timeout_cycles = 1;
+			else if (timeout_cycles > cycles_max)
+				timeout_cycles = cycles_max;
+		} else {
+			timeout_cycles = 0;
+		}
+	} else {
+		timeout_base_divisor = 0;
+		timeout_cycles = 0;
+	}
+
+	clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+				  timeout_base_divisor);
+
 	writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1);
-	writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2);
+	writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2);
 
 	return 0;
 }
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
 		}
 	}
 
+	device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+				 &bus->hw_timeout_ms);
+
 	/* Initialize the I2C adapter */
 	spin_lock_init(&bus->lock);
 	init_completion(&bus->cmd_complete);
Signature · 1 line
2.23.0
#3 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Thank you Jae for the patch. Let me play with it and will update back with my results later this week.

Cheers,

Tao

Quoted 171 lines
---
 drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---
 1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +73 -6
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
 #define ASPEED_I2CD_TIME_SCL_HIGH_MASK			GENMASK(19, 16)
 #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT			12
 #define ASPEED_I2CD_TIME_SCL_LOW_MASK			GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT	8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK	GENMASK(9, 8)
 #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK		GENMASK(3, 0)
 #define ASPEED_I2CD_TIME_SCL_REG_MAX			GENMASK(3, 0)
+
 /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL				0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT		0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK			GENMASK(4, 0)
 
 /* 0x0c : I2CD Interrupt Control Register &
  * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
  * These share bit definitions, so use the same values for the enable &
  * status bits.
  */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT		BIT(15)
 #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT			BIT(14)
 #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE		BIT(13)
 #define ASPEED_I2CD_INTR_SLAVE_MATCH			BIT(7)
@@ -96,8 +101,11 @@
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
 		 ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS					       \
+		ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
 #define ASPEED_I2CD_INTR_ALL						       \
-		(ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
+		(ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |		       \
+		 ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
 		 ASPEED_I2CD_INTR_BUS_RECOVER_DONE |			       \
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
 							   u32 divisor);
 	unsigned long			parent_clk_frequency;
 	u32				bus_frequency;
+	u32				hw_timeout_ms;
 	/* Transaction state. */
 	enum aspeed_i2c_master_state	master_state;
 	struct i2c_msg			*msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
 }
 
 #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+	if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+		return -EIO;
+
+	return 0;
+}
+
 static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 {
 	u32 command, irq_handled = 0;
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	if (!slave)
 		return 0;
 
+	if (aspeed_i2c_check_slave_error(irq_status)) {
+		dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+			irq_status);
+		irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
+		return irq_handled;
+	}
+
 	command = readl(bus->base + ASPEED_I2C_CMD_REG);
 
 	/* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
 	}
 }
 
-static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
 {
 	if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
 		return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	 * should clear the command queue effectively taking us back to the
 	 * INACTIVE state.
 	 */
-	ret = aspeed_i2c_is_irq_error(irq_status);
+	ret = aspeed_i2c_check_master_error(irq_status);
 	if (ret) {
-		dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+		dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
 			irq_status);
 		irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
 		if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
 /* precondition: bus.lock has been acquired. */
 static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 {
+	u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
 	u32 divisor, clk_reg_val;
 
 	divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 			ASPEED_I2CD_TIME_THDSTA_MASK |
 			ASPEED_I2CD_TIME_TACST_MASK);
 	clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);
+
+	if (bus->hw_timeout_ms) {
+		u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+			     ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+		u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+				ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+		timeout_base_divisor = 0;
+
+		do {
+			timeout_tick_us = 1000 * (16384 <<
+						  (timeout_base_divisor << 1)) /
+					  (bus->parent_clk_frequency / 1000);
+
+			if (timeout_base_divisor == div_max ||
+			    timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+			    bus->hw_timeout_ms * 1000)
+				break;
+		} while (timeout_base_divisor++ < div_max);
+
+		if (timeout_tick_us) {
+			timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+						      timeout_tick_us);
+			if (timeout_cycles == 0)
+				timeout_cycles = 1;
+			else if (timeout_cycles > cycles_max)
+				timeout_cycles = cycles_max;
+		} else {
+			timeout_cycles = 0;
+		}
+	} else {
+		timeout_base_divisor = 0;
+		timeout_cycles = 0;
+	}
+
+	clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+				  timeout_base_divisor);
+
 	writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1);
-	writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2);
+	writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2);
 
 	return 0;
 }
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
 		}
 	}
 
+	device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+				 &bus->hw_timeout_ms);
+
 	/* Initialize the I2C adapter */
 	spin_lock_init(&bus->lock);
 	init_completion(&bus->cmd_complete);
#4 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

In reply to Jae Hyun Yoo

Hi Jae,

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Cheers,

Tao

Quoted 171 lines
---
 drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---
 1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +73 -6
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
 #define ASPEED_I2CD_TIME_SCL_HIGH_MASK			GENMASK(19, 16)
 #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT			12
 #define ASPEED_I2CD_TIME_SCL_LOW_MASK			GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT	8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK	GENMASK(9, 8)
 #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK		GENMASK(3, 0)
 #define ASPEED_I2CD_TIME_SCL_REG_MAX			GENMASK(3, 0)
+
 /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL				0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT		0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK			GENMASK(4, 0)
 
 /* 0x0c : I2CD Interrupt Control Register &
  * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
  * These share bit definitions, so use the same values for the enable &
  * status bits.
  */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT		BIT(15)
 #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT			BIT(14)
 #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE		BIT(13)
 #define ASPEED_I2CD_INTR_SLAVE_MATCH			BIT(7)
@@ -96,8 +101,11 @@
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
 		 ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS					       \
+		ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
 #define ASPEED_I2CD_INTR_ALL						       \
-		(ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
+		(ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |		       \
+		 ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
 		 ASPEED_I2CD_INTR_BUS_RECOVER_DONE |			       \
 		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
 		 ASPEED_I2CD_INTR_ABNORMAL |				       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
 							   u32 divisor);
 	unsigned long			parent_clk_frequency;
 	u32				bus_frequency;
+	u32				hw_timeout_ms;
 	/* Transaction state. */
 	enum aspeed_i2c_master_state	master_state;
 	struct i2c_msg			*msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
 }
 
 #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+	if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+		return -EIO;
+
+	return 0;
+}
+
 static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 {
 	u32 command, irq_handled = 0;
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	if (!slave)
 		return 0;
 
+	if (aspeed_i2c_check_slave_error(irq_status)) {
+		dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+			irq_status);
+		irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
+		return irq_handled;
+	}
+
 	command = readl(bus->base + ASPEED_I2C_CMD_REG);
 
 	/* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
 	}
 }
 
-static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
 {
 	if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
 		return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	 * should clear the command queue effectively taking us back to the
 	 * INACTIVE state.
 	 */
-	ret = aspeed_i2c_is_irq_error(irq_status);
+	ret = aspeed_i2c_check_master_error(irq_status);
 	if (ret) {
-		dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+		dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
 			irq_status);
 		irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
 		if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
 /* precondition: bus.lock has been acquired. */
 static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 {
+	u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
 	u32 divisor, clk_reg_val;
 
 	divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
 			ASPEED_I2CD_TIME_THDSTA_MASK |
 			ASPEED_I2CD_TIME_TACST_MASK);
 	clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);
+
+	if (bus->hw_timeout_ms) {
+		u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+			     ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+		u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+				ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+		timeout_base_divisor = 0;
+
+		do {
+			timeout_tick_us = 1000 * (16384 <<
+						  (timeout_base_divisor << 1)) /
+					  (bus->parent_clk_frequency / 1000);
+
+			if (timeout_base_divisor == div_max ||
+			    timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+			    bus->hw_timeout_ms * 1000)
+				break;
+		} while (timeout_base_divisor++ < div_max);
+
+		if (timeout_tick_us) {
+			timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+						      timeout_tick_us);
+			if (timeout_cycles == 0)
+				timeout_cycles = 1;
+			else if (timeout_cycles > cycles_max)
+				timeout_cycles = cycles_max;
+		} else {
+			timeout_cycles = 0;
+		}
+	} else {
+		timeout_base_divisor = 0;
+		timeout_cycles = 0;
+	}
+
+	clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+				  timeout_base_divisor);
+
 	writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1);
-	writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2);
+	writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2);
 
 	return 0;
 }
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
 		}
 	}
 
+	device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+				 &bus->hw_timeout_ms);
+
 	/* Initialize the I2C adapter */
 	spin_lock_init(&bus->lock);
 	init_completion(&bus->cmd_complete);
#5 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

Quoted 73 lines
Quoted 73 lines
---
  drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---
  1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +19 -2
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
  #define ASPEED_I2CD_TIME_SCL_HIGH_MASK			GENMASK(19, 16)
  #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT			12
  #define ASPEED_I2CD_TIME_SCL_LOW_MASK			GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT	8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK	GENMASK(9, 8)
  #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK		GENMASK(3, 0)
  #define ASPEED_I2CD_TIME_SCL_REG_MAX			GENMASK(3, 0)
+
  /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL				0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT		0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK			GENMASK(4, 0)
 
  /* 0x0c : I2CD Interrupt Control Register &
   * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
   * These share bit definitions, so use the same values for the enable &
   * status bits.
   */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT		BIT(15)
  #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT			BIT(14)
  #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE		BIT(13)
  #define ASPEED_I2CD_INTR_SLAVE_MATCH			BIT(7)
@@ -96,8 +101,11 @@
  		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
  		 ASPEED_I2CD_INTR_ABNORMAL |				       \
  		 ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS					       \
+		ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
  #define ASPEED_I2CD_INTR_ALL						       \
-		(ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
+		(ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |		       \
+		 ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |			       \
  		 ASPEED_I2CD_INTR_BUS_RECOVER_DONE |			       \
  		 ASPEED_I2CD_INTR_SCL_TIMEOUT |				       \
  		 ASPEED_I2CD_INTR_ABNORMAL |				       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
  							   u32 divisor);
  	unsigned long			parent_clk_frequency;
  	u32				bus_frequency;
+	u32				hw_timeout_ms;
  	/* Transaction state. */
  	enum aspeed_i2c_master_state	master_state;
  	struct i2c_msg			*msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
  }
 
  #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+	if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+		return -EIO;
+
+	return 0;
+}
+
  static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
  {
  	u32 command, irq_handled = 0;
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
  	if (!slave)
  		return 0;
 
Quoted 5 lines

drivers/i2c/busses/i2c-aspeed.c static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status) · line 310

Quoted 5 lines
drivers/i2c/busses/i2c-aspeed.c +5 -0
+	if (aspeed_i2c_check_slave_error(irq_status)) {
+		dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+			irq_status);
+		irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
                aspeed_i2c_reset(bus);

I didn't add it in this patch because I wanted to avoid calling of this reset function from interrupt context but give it a try.

Thanks,
Jae

Quoted 94 lines
Quoted 94 lines
drivers/i2c/busses/i2c-aspeed.c +49 -4
+		return irq_handled;
+	}
+
  	command = readl(bus->base + ASPEED_I2C_CMD_REG);
 
  	/* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
  	}
  }
 
-static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
  {
  	if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
  		return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
  	 * should clear the command queue effectively taking us back to the
  	 * INACTIVE state.
  	 */
-	ret = aspeed_i2c_is_irq_error(irq_status);
+	ret = aspeed_i2c_check_master_error(irq_status);
  	if (ret) {
-		dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+		dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
  			irq_status);
  		irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
  		if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
  /* precondition: bus.lock has been acquired. */
  static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
  {
+	u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
  	u32 divisor, clk_reg_val;
 
  	divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
  			ASPEED_I2CD_TIME_THDSTA_MASK |
  			ASPEED_I2CD_TIME_TACST_MASK);
  	clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);
+
+	if (bus->hw_timeout_ms) {
+		u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+			     ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+		u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+				ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+		timeout_base_divisor = 0;
+
+		do {
+			timeout_tick_us = 1000 * (16384 <<
+						  (timeout_base_divisor << 1)) /
+					  (bus->parent_clk_frequency / 1000);
+
+			if (timeout_base_divisor == div_max ||
+			    timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+			    bus->hw_timeout_ms * 1000)
+				break;
+		} while (timeout_base_divisor++ < div_max);
+
+		if (timeout_tick_us) {
+			timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+						      timeout_tick_us);
+			if (timeout_cycles == 0)
+				timeout_cycles = 1;
+			else if (timeout_cycles > cycles_max)
+				timeout_cycles = cycles_max;
+		} else {
+			timeout_cycles = 0;
+		}
+	} else {
+		timeout_base_divisor = 0;
+		timeout_cycles = 0;
+	}
+
+	clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+				  timeout_base_divisor);
+
  	writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1);
-	writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2);
+	writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2);
 
  	return 0;
  }
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
  		}
  	}
 
+	device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+				 &bus->hw_timeout_ms);
+
  	/* Initialize the I2C adapter */
  	spin_lock_init(&bus->lock);
  	init_completion(&bus->cmd_complete);
#6 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Thanks,

Tao

Quoted 79 lines
Quoted 68 lines
Quoted 68 lines

---
  drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---

  1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +19 -2
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
   #define ASPEED_I2CD_TIME_SCL_HIGH_MASK            GENMASK(19, 16)
   #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT            12
   #define ASPEED_I2CD_TIME_SCL_LOW_MASK            GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT    8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK    GENMASK(9, 8)
   #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK        GENMASK(3, 0)
   #define ASPEED_I2CD_TIME_SCL_REG_MAX            GENMASK(3, 0)
+
   /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL                0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT        0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK            GENMASK(4, 0)
     /* 0x0c : I2CD Interrupt Control Register &
    * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
    * These share bit definitions, so use the same values for the enable &
    * status bits.
    */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT        BIT(15)
   #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT            BIT(14)
   #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE        BIT(13)
   #define ASPEED_I2CD_INTR_SLAVE_MATCH            BIT(7)
@@ -96,8 +101,11 @@
            ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
            ASPEED_I2CD_INTR_ABNORMAL |                       \
            ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS                           \
+        ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
   #define ASPEED_I2CD_INTR_ALL                               \
-        (ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
+        (ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |               \
+         ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
            ASPEED_I2CD_INTR_BUS_RECOVER_DONE |                   \
            ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
            ASPEED_I2CD_INTR_ABNORMAL |                       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
                                  u32 divisor);
       unsigned long            parent_clk_frequency;
       u32                bus_frequency;
+    u32                hw_timeout_ms;
       /* Transaction state. */
       enum aspeed_i2c_master_state    master_state;
       struct i2c_msg            *msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
   }
     #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+    if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+        return -EIO;
+
+    return 0;
+}
+
   static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
   {
       u32 command, irq_handled = 0;
Quoted 8 lines

drivers/i2c/busses/i2c-aspeed.c static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status) · line 309

Quoted 8 lines
drivers/i2c/busses/i2c-aspeed.c +4 -0
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
       if (!slave)
           return 0;
   +    if (aspeed_i2c_check_slave_error(irq_status)) {
+        dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+            irq_status);
+        irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+        bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;

                aspeed_i2c_reset(bus);

Quoted 5 lines

I didn't add it in this patch because I wanted to avoid calling of this reset function from interrupt context but give it a try.

Thanks,
Jae

I believe this will solve my problem, but let me test it and will share you results later.

Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
drivers/i2c/busses/i2c-aspeed.c +7 -2
+        return irq_handled;
+    }
+
       command = readl(bus->base + ASPEED_I2C_CMD_REG);
         /* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
       }
   }
   -static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
   {
       if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
           return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
        * should clear the command queue effectively taking us back to the
        * INACTIVE state.
        */
-    ret = aspeed_i2c_is_irq_error(irq_status);
+    ret = aspeed_i2c_check_master_error(irq_status);
       if (ret) {
-        dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+        dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
               irq_status);
           irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
           if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
   /* precondition: bus.lock has been acquired. */
   static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
   {
+    u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
       u32 divisor, clk_reg_val;
         divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)

              ASPEED_I2CD_TIME_THDSTA_MASK | ASPEED_I2CD_TIME_TACST_MASK); clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);

drivers/i2c/busses/i2c-aspeed.c +38 -0
+
+    if (bus->hw_timeout_ms) {
+        u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+                 ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+        u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+                ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+        timeout_base_divisor = 0;
+
+        do {
+            timeout_tick_us = 1000 * (16384 <<
+                          (timeout_base_divisor << 1)) /
+                      (bus->parent_clk_frequency / 1000);
+
+            if (timeout_base_divisor == div_max ||
+                timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+                bus->hw_timeout_ms * 1000)
+                break;
+        } while (timeout_base_divisor++ < div_max);
+
+        if (timeout_tick_us) {
+            timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+                              timeout_tick_us);
+            if (timeout_cycles == 0)
+                timeout_cycles = 1;
+            else if (timeout_cycles > cycles_max)
+                timeout_cycles = cycles_max;
+        } else {
+            timeout_cycles = 0;
+        }
+    } else {
+        timeout_base_divisor = 0;
+        timeout_cycles = 0;
+    }
+
+    clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+                  timeout_base_divisor);
+

      writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1); -    writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2); +    writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2); return 0;
  }

drivers/i2c/busses/i2c-aspeed.c +2 -0
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
           }
       }
   +    device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+                 &bus->hw_timeout_ms);
+
       /* Initialize the I2C adapter */
       spin_lock_init(&bus->lock);
       init_completion(&bus->cmd_complete);
#7 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Quoted 25 lines · Tao Ren
Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 8 lines
Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Oh, so to support the case as well, probably we need to add a flag for indicating recovery needs when a master xfer comes then it could forcibly recover and reset the bus even if the bus is idle. Can you please test that with making code changes? Unfortunately, I can't reproduce the case in my system.

Quoted 79 lines
Quoted 79 lines
Quoted 68 lines
Quoted 68 lines

---
   drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---
   1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +19 -2
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
   #define ASPEED_I2CD_TIME_SCL_HIGH_MASK            GENMASK(19, 16)
   #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT            12
   #define ASPEED_I2CD_TIME_SCL_LOW_MASK            GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT    8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK    GENMASK(9, 8)
   #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK        GENMASK(3, 0)
   #define ASPEED_I2CD_TIME_SCL_REG_MAX            GENMASK(3, 0)
+
   /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL                0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT        0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK            GENMASK(4, 0)
     /* 0x0c : I2CD Interrupt Control Register &
    * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
    * These share bit definitions, so use the same values for the enable &
    * status bits.
    */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT        BIT(15)
   #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT            BIT(14)
   #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE        BIT(13)
   #define ASPEED_I2CD_INTR_SLAVE_MATCH            BIT(7)
@@ -96,8 +101,11 @@
            ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
            ASPEED_I2CD_INTR_ABNORMAL |                       \
            ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS                           \
+        ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
   #define ASPEED_I2CD_INTR_ALL                               \
-        (ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
+        (ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |               \
+         ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
            ASPEED_I2CD_INTR_BUS_RECOVER_DONE |                   \
            ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
            ASPEED_I2CD_INTR_ABNORMAL |                       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
                                  u32 divisor);
       unsigned long            parent_clk_frequency;
       u32                bus_frequency;
+    u32                hw_timeout_ms;
       /* Transaction state. */
       enum aspeed_i2c_master_state    master_state;
       struct i2c_msg            *msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
   }
     #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+    if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+        return -EIO;
+
+    return 0;
+}
+
   static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
   {
       u32 command, irq_handled = 0;
Quoted 8 lines

drivers/i2c/busses/i2c-aspeed.c static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status) · line 309

Quoted 8 lines
drivers/i2c/busses/i2c-aspeed.c +4 -0
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
       if (!slave)
           return 0;
   +    if (aspeed_i2c_check_slave_error(irq_status)) {
+        dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+            irq_status);
+        irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+        bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;

                aspeed_i2c_reset(bus);

Quoted 7 lines
Quoted 5 lines

I didn't add it in this patch because I wanted to avoid calling of this reset function from interrupt context but give it a try.

Thanks,
Jae

I believe this will solve my problem, but let me test it and will share you results later.

Please share this result too. It would be useful to complete making this patch.

Thanks,
Jae

Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
drivers/i2c/busses/i2c-aspeed.c +48 -3
+        return irq_handled;
+    }
+
       command = readl(bus->base + ASPEED_I2C_CMD_REG);
         /* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
       }
   }
   -static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
   {
       if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
           return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
        * should clear the command queue effectively taking us back to the
        * INACTIVE state.
        */
-    ret = aspeed_i2c_is_irq_error(irq_status);
+    ret = aspeed_i2c_check_master_error(irq_status);
       if (ret) {
-        dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+        dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
               irq_status);
           irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
           if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
   /* precondition: bus.lock has been acquired. */
   static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
   {
+    u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
       u32 divisor, clk_reg_val;
         divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
               ASPEED_I2CD_TIME_THDSTA_MASK |
               ASPEED_I2CD_TIME_TACST_MASK);
       clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);
+
+    if (bus->hw_timeout_ms) {
+        u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+                 ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+        u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+                ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+        timeout_base_divisor = 0;
+
+        do {
+            timeout_tick_us = 1000 * (16384 <<
+                          (timeout_base_divisor << 1)) /
+                      (bus->parent_clk_frequency / 1000);
+
+            if (timeout_base_divisor == div_max ||
+                timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+                bus->hw_timeout_ms * 1000)
+                break;
+        } while (timeout_base_divisor++ < div_max);
+
+        if (timeout_tick_us) {
+            timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+                              timeout_tick_us);
+            if (timeout_cycles == 0)
+                timeout_cycles = 1;
+            else if (timeout_cycles > cycles_max)
+                timeout_cycles = cycles_max;
+        } else {
+            timeout_cycles = 0;
+        }
+    } else {
+        timeout_base_divisor = 0;
+        timeout_cycles = 0;
+    }
+
+    clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+                  timeout_base_divisor);
+
       writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1);
-    writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2);
+    writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2);
         return 0;
   }
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
           }
       }
   +    device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+                 &bus->hw_timeout_ms);
+
       /* Initialize the I2C adapter */
       spin_lock_init(&bus->lock);
       init_completion(&bus->cmd_complete);
#8 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Hi Jae,

Quoted 35 lines · Jae Hyun Yoo
Quoted 25 lines · Tao Ren
Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 8 lines
Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Quoted 5 lines

Oh, so to support the case as well, probably we need to add a flag for indicating recovery needs when a master xfer comes then it could forcibly recover and reset the bus even if the bus is idle. Can you please test that with making code changes? Unfortunately, I can't reproduce the case in my system.

Not sure if I understand it correctly, but given we already reset the bus in interrupt handler, the extra flag should not be needed?

Quoted 85 lines
Quoted 85 lines
Quoted 84 lines
Quoted 68 lines
Quoted 68 lines

---
   drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---

   1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +19 -2
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
    #define ASPEED_I2CD_TIME_SCL_HIGH_MASK            GENMASK(19, 16)
    #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT            12
    #define ASPEED_I2CD_TIME_SCL_LOW_MASK            GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT    8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK    GENMASK(9, 8)
    #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK        GENMASK(3, 0)
    #define ASPEED_I2CD_TIME_SCL_REG_MAX            GENMASK(3, 0)
+
    /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL                0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT        0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK            GENMASK(4, 0)
      /* 0x0c : I2CD Interrupt Control Register &
     * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
     * These share bit definitions, so use the same values for the enable &
     * status bits.
     */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT        BIT(15)
    #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT            BIT(14)
    #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE        BIT(13)
    #define ASPEED_I2CD_INTR_SLAVE_MATCH            BIT(7)
@@ -96,8 +101,11 @@
             ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
             ASPEED_I2CD_INTR_ABNORMAL |                       \
             ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS                           \
+        ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
    #define ASPEED_I2CD_INTR_ALL                               \
-        (ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
+        (ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |               \
+         ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
             ASPEED_I2CD_INTR_BUS_RECOVER_DONE |                   \
             ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
             ASPEED_I2CD_INTR_ABNORMAL |                       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
                                   u32 divisor);
        unsigned long            parent_clk_frequency;
        u32                bus_frequency;
+    u32                hw_timeout_ms;
        /* Transaction state. */
        enum aspeed_i2c_master_state    master_state;
        struct i2c_msg            *msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
    }
      #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+    if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+        return -EIO;
+
+    return 0;
+}
+
    static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
    {
        u32 command, irq_handled = 0;
Quoted 8 lines

drivers/i2c/busses/i2c-aspeed.c static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status) · line 309

Quoted 8 lines
drivers/i2c/busses/i2c-aspeed.c +4 -0
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
        if (!slave)
            return 0;
    +    if (aspeed_i2c_check_slave_error(irq_status)) {
+        dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+            irq_status);
+        irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+        bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;

                 aspeed_i2c_reset(bus);

I didn't add it in this patch because I wanted to avoid calling of this reset function from interrupt context but give it a try.

Thanks,
Jae

Quoted 7 lines
Quoted 1 line

I believe this will solve my problem, but let me test it and will share you results later.

Please share this result too. It would be useful to complete making this patch.

Thanks,
Jae

Adding aspeed_i2c_reset(bus) does fix my problem. Thank you Jae.

Cheers,

Tao

Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
drivers/i2c/busses/i2c-aspeed.c +7 -2
+        return irq_handled;
+    }
+
        command = readl(bus->base + ASPEED_I2C_CMD_REG);
          /* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
        }
    }
    -static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
    {
        if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
            return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
         * should clear the command queue effectively taking us back to the
         * INACTIVE state.
         */
-    ret = aspeed_i2c_is_irq_error(irq_status);
+    ret = aspeed_i2c_check_master_error(irq_status);
        if (ret) {
-        dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+        dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
                irq_status);
            irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
            if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
    /* precondition: bus.lock has been acquired. */
    static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
    {
+    u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
        u32 divisor, clk_reg_val;
          divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)

               ASPEED_I2CD_TIME_THDSTA_MASK | ASPEED_I2CD_TIME_TACST_MASK); clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);

drivers/i2c/busses/i2c-aspeed.c +38 -0
+
+    if (bus->hw_timeout_ms) {
+        u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+                 ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+        u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+                ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+        timeout_base_divisor = 0;
+
+        do {
+            timeout_tick_us = 1000 * (16384 <<
+                          (timeout_base_divisor << 1)) /
+                      (bus->parent_clk_frequency / 1000);
+
+            if (timeout_base_divisor == div_max ||
+                timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+                bus->hw_timeout_ms * 1000)
+                break;
+        } while (timeout_base_divisor++ < div_max);
+
+        if (timeout_tick_us) {
+            timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+                              timeout_tick_us);
+            if (timeout_cycles == 0)
+                timeout_cycles = 1;
+            else if (timeout_cycles > cycles_max)
+                timeout_cycles = cycles_max;
+        } else {
+            timeout_cycles = 0;
+        }
+    } else {
+        timeout_base_divisor = 0;
+        timeout_cycles = 0;
+    }
+
+    clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+                  timeout_base_divisor);
+

       writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1); -    writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2); +    writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2); return 0;
   }

drivers/i2c/busses/i2c-aspeed.c +2 -0
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
            }
        }
    +    device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+                 &bus->hw_timeout_ms);
+
        /* Initialize the I2C adapter */
        spin_lock_init(&bus->lock);
        init_completion(&bus->cmd_complete);
#9 lore raw

Re: [Potential Spoof] Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Quoted 38 lines · Tao Ren

Hi Jae,

Quoted 35 lines · Jae Hyun Yoo
Quoted 33 lines · Tao Ren
Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Quoted 7 lines
Quoted 5 lines

Oh, so to support the case as well, probably we need to add a flag for indicating recovery needs when a master xfer comes then it could forcibly recover and reset the bus even if the bus is idle. Can you please test that with making code changes? Unfortunately, I can't reproduce the case in my system.

Not sure if I understand it correctly, but given we already reset the bus in interrupt handler, the extra flag should not be needed?

One of my colleagues reminded me to enable bus auto release (I2CD00[17]) which could avoid explicit bus reset in interrupt handler.

Let me try and will update back soon.

Thanks,

Tao

Quoted 189 lines
Quoted 85 lines
Quoted 85 lines
Quoted 84 lines
Quoted 68 lines
Quoted 68 lines

---
   drivers/i2c/busses/i2c-aspeed.c | 79 ++++++++++++++++++++++++++++++---

   1 file changed, 73 insertions(+), 6 deletions(-)
drivers/i2c/busses/i2c-aspeed.c +19 -2
index 89317929bee4..92e1a249b393 100644
@@ -70,10 +70,14 @@
    #define ASPEED_I2CD_TIME_SCL_HIGH_MASK            GENMASK(19, 16)
    #define ASPEED_I2CD_TIME_SCL_LOW_SHIFT            12
    #define ASPEED_I2CD_TIME_SCL_LOW_MASK            GENMASK(15, 12)
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT    8
+#define ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK    GENMASK(9, 8)
    #define ASPEED_I2CD_TIME_BASE_DIVISOR_MASK        GENMASK(3, 0)
    #define ASPEED_I2CD_TIME_SCL_REG_MAX            GENMASK(3, 0)
+
    /* 0x08 : I2CD Clock and AC Timing Control Register #2 */
-#define ASPEED_NO_TIMEOUT_CTRL                0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT        0
+#define ASPEED_I2CD_TIMEOUT_CYCLES_MASK            GENMASK(4, 0)
      /* 0x0c : I2CD Interrupt Control Register &
     * 0x10 : I2CD Interrupt Status Register
@@ -81,6 +85,7 @@
     * These share bit definitions, so use the same values for the enable &
     * status bits.
     */
+#define ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT        BIT(15)
    #define ASPEED_I2CD_INTR_SDA_DL_TIMEOUT            BIT(14)
    #define ASPEED_I2CD_INTR_BUS_RECOVER_DONE        BIT(13)
    #define ASPEED_I2CD_INTR_SLAVE_MATCH            BIT(7)
@@ -96,8 +101,11 @@
             ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
             ASPEED_I2CD_INTR_ABNORMAL |                       \
             ASPEED_I2CD_INTR_ARBIT_LOSS)
+#define ASPEED_I2CD_INTR_SLAVE_ERRORS                           \
+        ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT
    #define ASPEED_I2CD_INTR_ALL                               \
-        (ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
+        (ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT |               \
+         ASPEED_I2CD_INTR_SDA_DL_TIMEOUT |                   \
             ASPEED_I2CD_INTR_BUS_RECOVER_DONE |                   \
             ASPEED_I2CD_INTR_SCL_TIMEOUT |                       \
             ASPEED_I2CD_INTR_ABNORMAL |                       \
@@ -176,6 +184,7 @@ struct aspeed_i2c_bus {
                                   u32 divisor);
        unsigned long            parent_clk_frequency;
        u32                bus_frequency;
+    u32                hw_timeout_ms;
        /* Transaction state. */
        enum aspeed_i2c_master_state    master_state;
        struct i2c_msg            *msgs;
@@ -276,6 +285,14 @@ static int aspeed_i2c_recover_bus(struct aspeed_i2c_bus *bus)
    }
      #if IS_ENABLED(CONFIG_I2C_SLAVE)
+static int aspeed_i2c_check_slave_error(u32 irq_status)
+{
+    if (irq_status & ASPEED_I2CD_INTR_SLAVE_INACTIVE_TIMEOUT)
+        return -EIO;
+
+    return 0;
+}
+
    static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
    {
        u32 command, irq_handled = 0;
Quoted 8 lines

drivers/i2c/busses/i2c-aspeed.c static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status) · line 309

Quoted 8 lines
drivers/i2c/busses/i2c-aspeed.c +4 -0
@@ -286,6 +303,14 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
        if (!slave)
            return 0;
    +    if (aspeed_i2c_check_slave_error(irq_status)) {
+        dev_dbg(bus->dev, "received slave error interrupt: 0x%08x\n",
+            irq_status);
+        irq_handled |= (irq_status & ASPEED_I2CD_INTR_SLAVE_ERRORS);
+        bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;

                 aspeed_i2c_reset(bus);

I didn't add it in this patch because I wanted to avoid calling of this reset function from interrupt context but give it a try.

Thanks,
Jae

Quoted 7 lines
Quoted 1 line

I believe this will solve my problem, but let me test it and will share you results later.

Please share this result too. It would be useful to complete making this patch.

Thanks,
Jae

Adding aspeed_i2c_reset(bus) does fix my problem. Thank you Jae.

Cheers,

Tao

Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
Quoted 89 lines
drivers/i2c/busses/i2c-aspeed.c +7 -2
+        return irq_handled;
+    }
+
        command = readl(bus->base + ASPEED_I2C_CMD_REG);
          /* Slave was requested, restart state machine. */
@@ -602,7 +627,7 @@ static void aspeed_i2c_next_msg_or_stop(struct aspeed_i2c_bus *bus)
        }
    }
    -static int aspeed_i2c_is_irq_error(u32 irq_status)
+static int aspeed_i2c_check_master_error(u32 irq_status)
    {
        if (irq_status & ASPEED_I2CD_INTR_ARBIT_LOSS)
            return -EAGAIN;
@@ -633,9 +658,9 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
         * should clear the command queue effectively taking us back to the
         * INACTIVE state.
         */
-    ret = aspeed_i2c_is_irq_error(irq_status);
+    ret = aspeed_i2c_check_master_error(irq_status);
        if (ret) {
-        dev_dbg(bus->dev, "received error interrupt: 0x%08x\n",
+        dev_dbg(bus->dev, "received master error interrupt: 0x%08x\n",
                irq_status);
            irq_handled |= (irq_status & ASPEED_I2CD_INTR_MASTER_ERRORS);
            if (bus->master_state != ASPEED_I2C_MASTER_INACTIVE) {
@@ -1194,6 +1219,7 @@ static u32 aspeed_i2c_25xx_get_clk_reg_val(struct device *dev, u32 divisor)
    /* precondition: bus.lock has been acquired. */
    static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)
    {
+    u32 timeout_base_divisor, timeout_tick_us, timeout_cycles;
        u32 divisor, clk_reg_val;
          divisor = DIV_ROUND_UP(bus->parent_clk_frequency, bus->bus_frequency);
@@ -1202,8 +1228,46 @@ static int aspeed_i2c_init_clk(struct aspeed_i2c_bus *bus)

               ASPEED_I2CD_TIME_THDSTA_MASK | ASPEED_I2CD_TIME_TACST_MASK); clk_reg_val |= bus->get_clk_reg_val(bus->dev, divisor);

drivers/i2c/busses/i2c-aspeed.c +38 -0
+
+    if (bus->hw_timeout_ms) {
+        u8 div_max = ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK >>
+                 ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_SHIFT;
+        u8 cycles_max = ASPEED_I2CD_TIMEOUT_CYCLES_MASK >>
+                ASPEED_I2CD_TIMEOUT_CYCLES_SHIFT;
+
+        timeout_base_divisor = 0;
+
+        do {
+            timeout_tick_us = 1000 * (16384 <<
+                          (timeout_base_divisor << 1)) /
+                      (bus->parent_clk_frequency / 1000);
+
+            if (timeout_base_divisor == div_max ||
+                timeout_tick_us * ASPEED_I2CD_TIMEOUT_CYCLES_MASK >=
+                bus->hw_timeout_ms * 1000)
+                break;
+        } while (timeout_base_divisor++ < div_max);
+
+        if (timeout_tick_us) {
+            timeout_cycles = DIV_ROUND_UP(bus->hw_timeout_ms * 1000,
+                              timeout_tick_us);
+            if (timeout_cycles == 0)
+                timeout_cycles = 1;
+            else if (timeout_cycles > cycles_max)
+                timeout_cycles = cycles_max;
+        } else {
+            timeout_cycles = 0;
+        }
+    } else {
+        timeout_base_divisor = 0;
+        timeout_cycles = 0;
+    }
+
+    clk_reg_val |= FIELD_PREP(ASPEED_I2CD_TIME_TIMEOUT_BASE_DIVISOR_MASK,
+                  timeout_base_divisor);
+

       writel(clk_reg_val, bus->base + ASPEED_I2C_AC_TIMING_REG1); -    writel(ASPEED_NO_TIMEOUT_CTRL, bus->base + ASPEED_I2C_AC_TIMING_REG2); +    writel(timeout_cycles, bus->base + ASPEED_I2C_AC_TIMING_REG2); return 0;
   }

drivers/i2c/busses/i2c-aspeed.c +2 -0
@@ -1404,6 +1468,9 @@ static int aspeed_i2c_probe_bus(struct platform_device *pdev)
            }
        }
    +    device_property_read_u32(&pdev->dev, "aspeed,hw-timeout-ms",
+                 &bus->hw_timeout_ms);
+
        /* Initialize the I2C adapter */
        spin_lock_init(&bus->lock);
        init_completion(&bus->cmd_complete);
#10 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Hi Jae,

Quoted 45 lines · Tao Ren
Quoted 44 lines · Tao Ren

Hi Jae,

Quoted 40 lines · Jae Hyun Yoo
Quoted 25 lines · Tao Ren
Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 8 lines
Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Oh, so to support the case as well, probably we need to add a flag for indicating recovery needs when a master xfer comes then it could forcibly recover and reset the bus even if the bus is idle. Can you please test that with making code changes? Unfortunately, I can't reproduce the case in my system.

Quoted 5 lines
Quoted 1 line

Not sure if I understand it correctly, but given we already reset the bus in interrupt handler, the extra flag should not be needed?

One of my colleagues reminded me to enable bus auto release (I2CD00[17]) which could avoid explicit bus reset in interrupt handler.

Let me try and will update back soon.

Let me summarize my testing result on Minipack BMC:

BMC i2c controller cannot exit slave state by enabling slave_inactive_timeout interrupt only: in this case, i2c controller stays in SRXD state and slave_inactive_timeout interrupt will be generated again and again until bus is explicitly reset.

I've tried below 2 approaches and both works on my platform:
- calling aspeed_i2c_reset() in interrupt handler when slave_inacive_timeout interrupt is delivered.
- enabling bus auto release feature (I2CD00[17]) in probe function.

Cheers,

Tao

#11 lore raw

Re: [PATCH dev-5.2 2/2] i2c: aspeed: add slave inactive timeout support

Hi Tao,

Quoted 54 lines · Tao Ren

Hi Jae,

Quoted 45 lines · Tao Ren
Quoted 44 lines · Tao Ren

Hi Jae,

Quoted 40 lines · Jae Hyun Yoo
Quoted 25 lines · Tao Ren
Quoted 24 lines · Jae Hyun Yoo

Hi Tao,

Quoted 8 lines · Tao Ren

Hi Jae,

Quoted 5 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 1 line

I have a generic question on the patch: is it good enough to just enable slave_inactive_timeout and acknowledge the interrupt when it happens? Or do we need to reset the i2c controller to bring it out of slave state?

I'm asking this because I hit an issue on my Minipack BMC where the slave_inactive_timeout interrupt was generated once every ~0.1 seconds (I set aspeed,hw-timeout-ms to 1000), and a few seconds later my BMC was rebooted (I guess watchdog timed out). Given the issue happened only once, I don't have chance to collect more information (such as why the repetitive interrupt was generated, why BMC rebooted, and etc.). Will share more if I met the problem again.

Basic concept of this implementation is setting the slave state of driver to ASPEED_I2C_SLAVE_INACTIVE to avoid calling of aspeed_i2c_reset() directly from interrupt context. Instead, when a master xfer happens after that, it will try bus recovery through aspeed_i2c_recover_bus() and it will call aspeed_i2c_reset() if needed.

Quoted 8 lines
Quoted 2 lines

If this patch doesn't work in your case, test it again after adding one line code into this driver. See below.

If bus_reset is really needed in this case, then I'd prefer to do it immediately or in a threaded_irq_handler if it's bad idea to call aspeed_i2c_reset() in hardware interrupt context. The reasons being:

1) we don't know when userspace starts next master transfer.
2) aspeed_i2c_recover_bus() is not triggered in aspeed_i2c_master_xfer() in my environment because bus is "idle" (I2CD14[16] == 0).

Oh, so to support the case as well, probably we need to add a flag for indicating recovery needs when a master xfer comes then it could forcibly recover and reset the bus even if the bus is idle. Can you please test that with making code changes? Unfortunately, I can't reproduce the case in my system.

Quoted 5 lines
Quoted 1 line

Not sure if I understand it correctly, but given we already reset the bus in interrupt handler, the extra flag should not be needed?

One of my colleagues reminded me to enable bus auto release (I2CD00[17]) which could avoid explicit bus reset in interrupt handler.

Let me try and will update back soon.

Quoted 7 lines

Let me summarize my testing result on Minipack BMC:

BMC i2c controller cannot exit slave state by enabling slave_inactive_timeout interrupt only: in this case, i2c controller stays in SRXD state and slave_inactive_timeout interrupt will be generated again and again until bus is explicitly reset.

I've tried below 2 approaches and both works on my platform:
- calling aspeed_i2c_reset() in interrupt handler when slave_inacive_timeout interrupt is delivered.
- enabling bus auto release feature (I2CD00[17]) in probe function.

Thanks a lot for your testing result and suggestions. It seems that the latter way - enabling bus auto release would be better for this implementation.

I'll submit a next spin today.

Cheers,
Jae

maintainer
#12 lore raw
In reply to Jae Hyun Yoo

Hi Jae,

Quoted 8 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Cheers,

Joel

Quoted 12 lines

Jae Hyun Yoo (2):
dt-bindings: i2c: aspeed: add hardware timeout support
i2c: aspeed: add slave inactive timeout support

 .../devicetree/bindings/i2c/i2c-aspeed.txt    |  2 +
 drivers/i2c/busses/i2c-aspeed.c               | 79 +++++++++++++++++--
 2 files changed, 75 insertions(+), 6 deletions(-)

--
2.23.0

#13 lore raw

Hi Joel,

Quoted 9 lines · Joel Stanley

Hi Jae,

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 2 lines

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

Thanks,
Jae

Quoted 17 lines

Cheers,

Joel

Quoted 12 lines

Jae Hyun Yoo (2):
dt-bindings: i2c: aspeed: add hardware timeout support
i2c: aspeed: add slave inactive timeout support

  .../devicetree/bindings/i2c/i2c-aspeed.txt    |  2 +
  drivers/i2c/busses/i2c-aspeed.c               | 79 +++++++++++++++++--
  2 files changed, 75 insertions(+), 6 deletions(-)

--
2.23.0

#14 lore raw
Quoted 12 lines · Jae Hyun Yoo

Hi Joel,

Quoted 9 lines · Joel Stanley

Hi Jae,

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Cheers,

Andrew

#15 lore raw

Hi Andrew,

Quoted 19 lines · Andrew Jeffery
Quoted 18 lines · Jae Hyun Yoo

Hi Joel,

Quoted 14 lines · Joel Stanley

Hi Jae,

Quoted 8 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 6 lines
Quoted 2 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Thanks,
Jae

#16 lore raw
Quoted 29 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 19 lines · Andrew Jeffery
Quoted 18 lines · Jae Hyun Yoo

Hi Joel,

Quoted 14 lines · Joel Stanley

Hi Jae,

Quoted 8 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 6 lines
Quoted 2 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Quoted 5 lines

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Ah, didn't realise it was Guenter that ran into it. We have some changes[1] in Cedric's aspeed-4.2 qemu tree - do you mind testing it out? If those patches resolve the issue Maybe we could point Guenter at that tree, though really we should get the fixes upstream so this isn't an issue.

[1] https://github.com/legoater/qemu/compare/59dda66ab756e52e6a9c1ef89660d30b3769f63c...aspeed-4.2

Cheers,

Andrew

#17 lore raw
Quoted 38 lines · Andrew Jeffery
Quoted 29 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 19 lines · Andrew Jeffery
Quoted 18 lines · Jae Hyun Yoo

Hi Joel,

Quoted 14 lines · Joel Stanley

Hi Jae,

Quoted 8 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 6 lines
Quoted 2 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Quoted 5 lines

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Quoted 7 lines

Ah, didn't realise it was Guenter that ran into it. We have some changes[1] in Cedric's aspeed-4.2 qemu tree - do you mind testing it out? If those patches resolve the issue Maybe we could point Guenter at that tree, though really we should get the fixes upstream so this isn't an issue.

[1] https://github.com/legoater/qemu/compare/59dda66ab756e52e6a9c1ef89660d30b3769f63c...aspeed-4.2

Okay. I'll give it a try.

Thanks,
Jae

#18 lore raw

Hi Andrew,

Quoted 45 lines · Jae Hyun Yoo
Quoted 44 lines · Andrew Jeffery
Quoted 29 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 19 lines · Andrew Jeffery
Quoted 18 lines · Jae Hyun Yoo

Hi Joel,

Quoted 14 lines · Joel Stanley

Hi Jae,

Quoted 8 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 6 lines
Quoted 2 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Quoted 6 lines

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Ah, didn't realise it was Guenter that ran into it. We have some changes[1] in Cedric's aspeed-4.2 qemu tree - do you mind testing it out? If those patches resolve the issue Maybe we could point Guenter at that tree, though really we should get the fixes upstream so this isn't an issue.

Quoted 5 lines
Quoted 3 lines

[1] https://github.com/legoater/qemu/compare/59dda66ab756e52e6a9c1ef89660d30b3769f63c...aspeed-4.2

Okay. I'll give it a try.

I've tested I2C buffer mode support in QEMU using:
git://github.com/legoater/qemu.git
SRCBRANCH = "aspeed-4.2"
SRCREV = "1b31d645c448858eb7d11d463a4cb77df0ee7923"

Checked that I2C buffer mode works on the latest QEMU H/W model. Thanks Cedric, Eddie for the H/W model implementation.

I'll submit all I2C backlog patches into linux upstream.

Cheers,
Jae

#19 lore raw
Quoted 62 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 59 lines · Jae Hyun Yoo
Quoted 57 lines · Andrew Jeffery
Quoted 35 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 23 lines · Andrew Jeffery
Quoted 22 lines · Jae Hyun Yoo

Hi Joel,

Quoted 18 lines · Joel Stanley

Hi Jae,

Quoted 11 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles
a bus in the middle of a transaction, I2C hardware hangs in slave state
and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature
also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 8 lines
Quoted 3 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU
H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Quoted 6 lines

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Ah, didn't realise it was Guenter that ran into it. We have some changes[1] in
Cedric's aspeed-4.2 qemu tree - do you mind testing it out? If those patches
resolve the issue Maybe we could point Guenter at that tree, though really we
should get the fixes upstream so this isn't an issue.

[1]
https://github.com/legoater/qemu/compare/59dda66ab756e52e6a9c1ef89660d30b3769f63c...aspeed-4.2

Quoted 6 lines
Quoted 1 line

Okay. I'll give it a try.

I've tested I2C buffer mode support in QEMU using:
git://github.com/legoater/qemu.git
SRCBRANCH = "aspeed-4.2"
SRCREV = "1b31d645c448858eb7d11d463a4cb77df0ee7923"

This looks like changes to the qemu bitbake recipe? Have you integrated openbmc/qemu into Yocto's runqemu infrastructure, or were you just using bitbake to build qemu? I'd love to see patches if you've done the former :)

Andrew

#20 lore raw
Quoted 72 lines · Andrew Jeffery
Quoted 62 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 59 lines · Jae Hyun Yoo
Quoted 57 lines · Andrew Jeffery
Quoted 35 lines · Jae Hyun Yoo

Hi Andrew,

Quoted 23 lines · Andrew Jeffery
Quoted 22 lines · Jae Hyun Yoo

Hi Joel,

Quoted 18 lines · Joel Stanley

Hi Jae,

Quoted 11 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles
a bus in the middle of a transaction, I2C hardware hangs in slave state
and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

By applying this change, SDA data-low and SCL clock-low timeout feature
also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Quoted 8 lines
Quoted 3 lines

Thanks for the reminding. I can't send the RFC patches yet because QEMU
H/W model isn't ready yet. I'm still waiting for the fix from Cedric.

QEMU shouldn't be preventing you from sending patches upstream, rather it prevents us from enabling the buffer mode support in the OpenBMC kernel tree. You should be sending all patches upstream as early as possible.

Quoted 6 lines

I met a QEMU issue when I was upstreaming a patch set last year:
https://lists.ozlabs.org/pipermail/linux-aspeed/2018-September/000750.html

If OpenBMC community accepts the QEMU issue, I can submit the RFC patches to upstream. Will submit the patch set soon to linux tree.

Ah, didn't realise it was Guenter that ran into it. We have some changes[1] in
Cedric's aspeed-4.2 qemu tree - do you mind testing it out? If those patches
resolve the issue Maybe we could point Guenter at that tree, though really we
should get the fixes upstream so this isn't an issue.

[1]
https://github.com/legoater/qemu/compare/59dda66ab756e52e6a9c1ef89660d30b3769f63c...aspeed-4.2

Quoted 6 lines
Quoted 1 line

Okay. I'll give it a try.

I've tested I2C buffer mode support in QEMU using:
git://github.com/legoater/qemu.git
SRCBRANCH = "aspeed-4.2"
SRCREV = "1b31d645c448858eb7d11d463a4cb77df0ee7923"

Quoted 7 lines

This looks like changes to the qemu bitbake recipe? Have you integrated openbmc/qemu into Yocto's runqemu infrastructure, or were you just using bitbake to build qemu? I'd love to see patches if you've done the former :)

Andrew

No, I just made this change using a bbappend in my local yocto tree. The source revision on the branch worked well.

Jae

#21 lore raw
In reply to Joel Stanley
Quoted 10 lines · Joel Stanley

Hi Jae,

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 2 lines

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Joel, thanks for keeping track of this!

Sorry, for not keeping up with my emails, I don't think I have time to continue to maintain this.

However, I don't want to hand this off in a bad state. I will try to get up to date on all the emails in the coming weeks.

Jae, can you start sending things upstream as Joel suggested?

#22 lore raw
Quoted 19 lines · Brendan Higgins
Quoted 10 lines · Joel Stanley

Hi Jae,

Quoted 6 lines · Jae Hyun Yoo

In case of multi-master environment, if a peer master incorrectly handles a bus in the middle of a transaction, I2C hardware hangs in slave state and it can't escape from the slave state, so this commit adds slave inactive timeout support to recover the bus in the case.

Quoted 5 lines
Quoted 2 lines

By applying this change, SDA data-low and SCL clock-low timeout feature also could be enabled which was disabled previously.

Please consider sending your RFC patches to the upstream list. You have a big backlog of patches now.

Joel, thanks for keeping track of this!

Quoted 7 lines

Sorry, for not keeping up with my emails, I don't think I have time to continue to maintain this.

However, I don't want to hand this off in a bad state. I will try to get up to date on all the emails in the coming weeks.

Jae, can you start sending things upstream as Joel suggested?

Sure, I'll start upstreaming after taking some tests on Cedric's QEMU patch.

Thanks,
Jae