diff mbox series

[v3,1/2] i2c: aspeed: Handle the coalesced stop conditions with the start conditions.

Message ID 20231208033142.1673232-2-quan@os.amperecomputing.com
State New
Headers show
Series i2c: aspeed: Late ack Tx done irqs and handle coalesced start with stop conditions | expand

Commit Message

Quan Nguyen Dec. 8, 2023, 3:31 a.m. UTC
Some masters may drive the transfers with low enough latency between
the nak/stop phase of the current command and the start/address phase
of the following command that the interrupts are coalesced by the
time we process them.
Handle the stop conditions before processing SLAVE_MATCH to fix the
complaints that sometimes occur below.

"aspeed-i2c-bus 1e78a040.i2c-bus: irq handled != irq. Expected
0x00000086, but was 0x00000084"

Fixes: f9eb91350bb2 ("i2c: aspeed: added slave support for Aspeed I2C driver")
Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com>
---
v3:
  + Change to handle the coalesced stop condition with the start
conditions                                                            [Andrew]
  + Revised commit message                                              [Quan]

v2:
  + Split to separate series                                            [Joel]
  + Added the Fixes line                                                [Joel]
  + Revised commit message                                              [Quan]

v1:
  + First introduced in
https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/
---
 drivers/i2c/busses/i2c-aspeed.c | 47 ++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 16 deletions(-)

Comments

Andrew Jeffery Dec. 8, 2023, 3:56 a.m. UTC | #1
On Fri, 2023-12-08 at 10:31 +0700, Quan Nguyen wrote:
> Some masters may drive the transfers with low enough latency between
> the nak/stop phase of the current command and the start/address phase
> of the following command that the interrupts are coalesced by the
> time we process them.
> Handle the stop conditions before processing SLAVE_MATCH to fix the
> complaints that sometimes occur below.
> 
> "aspeed-i2c-bus 1e78a040.i2c-bus: irq handled != irq. Expected
> 0x00000086, but was 0x00000084"
> 
> Fixes: f9eb91350bb2 ("i2c: aspeed: added slave support for Aspeed I2C driver")
> Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com>
> ---
> v3:
>   + Change to handle the coalesced stop condition with the start
> conditions                                                            [Andrew]
>   + Revised commit message                                              [Quan]
> 
> v2:
>   + Split to separate series                                            [Joel]
>   + Added the Fixes line                                                [Joel]
>   + Revised commit message                                              [Quan]
> 
> v1:
>   + First introduced in
> https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/
> ---
>  drivers/i2c/busses/i2c-aspeed.c | 47 ++++++++++++++++++++++-----------
>  1 file changed, 31 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c
> index 28e2a5fc4528..1c2a4f4c4e1b 100644
> --- a/drivers/i2c/busses/i2c-aspeed.c
> +++ b/drivers/i2c/busses/i2c-aspeed.c
> @@ -249,18 +249,45 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
>  	if (!slave)
>  		return 0;
>  
> -	command = readl(bus->base + ASPEED_I2C_CMD_REG);
> +	/*
> +	 * Handle stop conditions early, prior to SLAVE_MATCH. Some masters may drive
> +	 * transfers with low enough latency between the nak/stop phase of the current
> +	 * command and the start/address phase of the following command that the
> +	 * interrupts are coalesced by the time we process them.
> +	 */
> +	if (irq_status & ASPEED_I2CD_INTR_NORMAL_STOP) {
> +		irq_handled |= ASPEED_I2CD_INTR_NORMAL_STOP;
> +		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
> +	}
>  
> -	/* Slave was requested, restart state machine. */
> +	if (irq_status & ASPEED_I2CD_INTR_TX_NAK &&
> +	    bus->slave_state == ASPEED_I2C_SLAVE_READ_PROCESSED) {
> +		irq_handled |= ASPEED_I2CD_INTR_TX_NAK;
> +		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
> +	}
> +
> +	/* Propagate any stop conditions to the slave implementation. */
> +	if (bus->slave_state == ASPEED_I2C_SLAVE_STOP) {
> +		i2c_slave_event(slave, I2C_SLAVE_STOP, &value);
> +		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
> +	}
> +	/*

If there's a reason to do a v4 then an extra empty line above the
comment would be nice. But let's not get hung up on that if everyone
else is happy.

Thanks for the fixes!

Reviewed-by: Andrew Jeffery <andrew@codeconstruct.com.au>

> +	 * Now that we've dealt with any potentially coalesced stop conditions,
> +	 * address any start conditions.
> +	 */
>  	if (irq_status & ASPEED_I2CD_INTR_SLAVE_MATCH) {
>  		irq_handled |= ASPEED_I2CD_INTR_SLAVE_MATCH;
>  		bus->slave_state = ASPEED_I2C_SLAVE_START;
>  	}
>
Andi Shyti Dec. 9, 2023, 8:28 p.m. UTC | #2
Hi Quan,

On Fri, Dec 08, 2023 at 10:31:41AM +0700, Quan Nguyen wrote:
> Some masters may drive the transfers with low enough latency between
> the nak/stop phase of the current command and the start/address phase
> of the following command that the interrupts are coalesced by the
> time we process them.
> Handle the stop conditions before processing SLAVE_MATCH to fix the
> complaints that sometimes occur below.
> 
> "aspeed-i2c-bus 1e78a040.i2c-bus: irq handled != irq. Expected
> 0x00000086, but was 0x00000084"
> 
> Fixes: f9eb91350bb2 ("i2c: aspeed: added slave support for Aspeed I2C driver")
> Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com>

Reviewed-by: Andi Shyti <andi.shyti@kernel.org>

Thanks,
Andi
Quan Nguyen Dec. 11, 2023, 4:08 a.m. UTC | #3
On 08/12/2023 10:56, Andrew Jeffery wrote:
> On Fri, 2023-12-08 at 10:31 +0700, Quan Nguyen wrote:
>> Some masters may drive the transfers with low enough latency between
>> the nak/stop phase of the current command and the start/address phase
>> of the following command that the interrupts are coalesced by the
>> time we process them.
>> Handle the stop conditions before processing SLAVE_MATCH to fix the
>> complaints that sometimes occur below.
>>
>> "aspeed-i2c-bus 1e78a040.i2c-bus: irq handled != irq. Expected
>> 0x00000086, but was 0x00000084"
>>
>> Fixes: f9eb91350bb2 ("i2c: aspeed: added slave support for Aspeed I2C driver")
>> Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com>
>> ---
>> v3:
>>    + Change to handle the coalesced stop condition with the start
>> conditions                                                            [Andrew]
>>    + Revised commit message                                              [Quan]
>>
>> v2:
>>    + Split to separate series                                            [Joel]
>>    + Added the Fixes line                                                [Joel]
>>    + Revised commit message                                              [Quan]
>>
>> v1:
>>    + First introduced in
>> https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/
>> ---
>>   drivers/i2c/busses/i2c-aspeed.c | 47 ++++++++++++++++++++++-----------
>>   1 file changed, 31 insertions(+), 16 deletions(-)
>>
>> diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c
>> index 28e2a5fc4528..1c2a4f4c4e1b 100644
>> --- a/drivers/i2c/busses/i2c-aspeed.c
>> +++ b/drivers/i2c/busses/i2c-aspeed.c
>> @@ -249,18 +249,45 @@ static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
>>   	if (!slave)
>>   		return 0;
>>   
>> -	command = readl(bus->base + ASPEED_I2C_CMD_REG);
>> +	/*
>> +	 * Handle stop conditions early, prior to SLAVE_MATCH. Some masters may drive
>> +	 * transfers with low enough latency between the nak/stop phase of the current
>> +	 * command and the start/address phase of the following command that the
>> +	 * interrupts are coalesced by the time we process them.
>> +	 */
>> +	if (irq_status & ASPEED_I2CD_INTR_NORMAL_STOP) {
>> +		irq_handled |= ASPEED_I2CD_INTR_NORMAL_STOP;
>> +		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
>> +	}
>>   
>> -	/* Slave was requested, restart state machine. */
>> +	if (irq_status & ASPEED_I2CD_INTR_TX_NAK &&
>> +	    bus->slave_state == ASPEED_I2C_SLAVE_READ_PROCESSED) {
>> +		irq_handled |= ASPEED_I2CD_INTR_TX_NAK;
>> +		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
>> +	}
>> +
>> +	/* Propagate any stop conditions to the slave implementation. */
>> +	if (bus->slave_state == ASPEED_I2C_SLAVE_STOP) {
>> +		i2c_slave_event(slave, I2C_SLAVE_STOP, &value);
>> +		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
>> +	}
>> +	/*
> 
> If there's a reason to do a v4 then an extra empty line above the
> comment would be nice. But let's not get hung up on that if everyone
> else is happy.
> 
> Thanks for the fixes!
> 
> Reviewed-by: Andrew Jeffery <andrew@codeconstruct.com.au>
> 

Thanks Andrew,

I'll add your Reviewed-by in my v4 with that extra empty lime before the 
comment.

Thanks for the review
- Quan
Quan Nguyen Dec. 11, 2023, 4:09 a.m. UTC | #4
On 10/12/2023 03:28, Andi Shyti wrote:
> Hi Quan,
> 
> On Fri, Dec 08, 2023 at 10:31:41AM +0700, Quan Nguyen wrote:
>> Some masters may drive the transfers with low enough latency between
>> the nak/stop phase of the current command and the start/address phase
>> of the following command that the interrupts are coalesced by the
>> time we process them.
>> Handle the stop conditions before processing SLAVE_MATCH to fix the
>> complaints that sometimes occur below.
>>
>> "aspeed-i2c-bus 1e78a040.i2c-bus: irq handled != irq. Expected
>> 0x00000086, but was 0x00000084"
>>
>> Fixes: f9eb91350bb2 ("i2c: aspeed: added slave support for Aspeed I2C driver")
>> Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com>
> 
> Reviewed-by: Andi Shyti <andi.shyti@kernel.org>
> 

Thanks Andi,
I'll add your Reviewed-by in v4

Thanks,
- Quan
diff mbox series

Patch

diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c
index 28e2a5fc4528..1c2a4f4c4e1b 100644
--- a/drivers/i2c/busses/i2c-aspeed.c
+++ b/drivers/i2c/busses/i2c-aspeed.c
@@ -249,18 +249,45 @@  static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 	if (!slave)
 		return 0;
 
-	command = readl(bus->base + ASPEED_I2C_CMD_REG);
+	/*
+	 * Handle stop conditions early, prior to SLAVE_MATCH. Some masters may drive
+	 * transfers with low enough latency between the nak/stop phase of the current
+	 * command and the start/address phase of the following command that the
+	 * interrupts are coalesced by the time we process them.
+	 */
+	if (irq_status & ASPEED_I2CD_INTR_NORMAL_STOP) {
+		irq_handled |= ASPEED_I2CD_INTR_NORMAL_STOP;
+		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
+	}
 
-	/* Slave was requested, restart state machine. */
+	if (irq_status & ASPEED_I2CD_INTR_TX_NAK &&
+	    bus->slave_state == ASPEED_I2C_SLAVE_READ_PROCESSED) {
+		irq_handled |= ASPEED_I2CD_INTR_TX_NAK;
+		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
+	}
+
+	/* Propagate any stop conditions to the slave implementation. */
+	if (bus->slave_state == ASPEED_I2C_SLAVE_STOP) {
+		i2c_slave_event(slave, I2C_SLAVE_STOP, &value);
+		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
+	}
+	/*
+	 * Now that we've dealt with any potentially coalesced stop conditions,
+	 * address any start conditions.
+	 */
 	if (irq_status & ASPEED_I2CD_INTR_SLAVE_MATCH) {
 		irq_handled |= ASPEED_I2CD_INTR_SLAVE_MATCH;
 		bus->slave_state = ASPEED_I2C_SLAVE_START;
 	}
 
-	/* Slave is not currently active, irq was for someone else. */
+	/*
+	 * If the slave has been stopped and not started then slave interrupt
+	 * handling is complete.
+	 */
 	if (bus->slave_state == ASPEED_I2C_SLAVE_INACTIVE)
 		return irq_handled;
 
+	command = readl(bus->base + ASPEED_I2C_CMD_REG);
 	dev_dbg(bus->dev, "slave irq status 0x%08x, cmd 0x%08x\n",
 		irq_status, command);
 
@@ -279,17 +306,6 @@  static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 		irq_handled |= ASPEED_I2CD_INTR_RX_DONE;
 	}
 
-	/* Slave was asked to stop. */
-	if (irq_status & ASPEED_I2CD_INTR_NORMAL_STOP) {
-		irq_handled |= ASPEED_I2CD_INTR_NORMAL_STOP;
-		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
-	}
-	if (irq_status & ASPEED_I2CD_INTR_TX_NAK &&
-	    bus->slave_state == ASPEED_I2C_SLAVE_READ_PROCESSED) {
-		irq_handled |= ASPEED_I2CD_INTR_TX_NAK;
-		bus->slave_state = ASPEED_I2C_SLAVE_STOP;
-	}
-
 	switch (bus->slave_state) {
 	case ASPEED_I2C_SLAVE_READ_REQUESTED:
 		if (unlikely(irq_status & ASPEED_I2CD_INTR_TX_ACK))
@@ -324,8 +340,7 @@  static u32 aspeed_i2c_slave_irq(struct aspeed_i2c_bus *bus, u32 irq_status)
 		i2c_slave_event(slave, I2C_SLAVE_WRITE_RECEIVED, &value);
 		break;
 	case ASPEED_I2C_SLAVE_STOP:
-		i2c_slave_event(slave, I2C_SLAVE_STOP, &value);
-		bus->slave_state = ASPEED_I2C_SLAVE_INACTIVE;
+		/* Stop event handling is done early. Unreachable. */
 		break;
 	case ASPEED_I2C_SLAVE_START:
 		/* Slave was just started. Waiting for the next event. */;