[RFC PATCH i2c-next 0/2] i2c: aspeed: Add bus idle waiting logic for multi-master use cases

[RFC PATCH i2c-next 0/2] i2c: aspeed: Add bus idle waiting logic for multi-master use cases

[Jae Hyun Yoo]

In multi-master environment, this driver side master cannot know
exactly when peer master sends data to this side master so a case
can be happened that this master tries to send data through the
master_xfer function but slave data from peer master is still
being processed by this driver.

To prevent state corruption in the case, this patch adds checking
if any slave operation is ongoing and it wait up to the timeout
duration before starting a master_xfer operation.

Please review this patch set.

Thanks,

-Jae

Jae Hyun Yoo (2):
  dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting
  i2c: aspeed: Add bus idle waiting logic for multi-master use cases

 .../devicetree/bindings/i2c/i2c-aspeed.txt    | 10 ++-
 drivers/i2c/busses/i2c-aspeed.c               | 70 +++++++++++++++----
 2 files changed, 62 insertions(+), 18 deletions(-)

-- 
2.18.0

[RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

This commit adds 'idle-wait-timeout-ms' setting which can be used
for bus idle waiting logic in multi-master environment.

Signed-off-by: Jae Hyun Yoo <jae.hyun.yoo@linux.intel.com>
---
 Documentation/devicetree/bindings/i2c/i2c-aspeed.txt | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/Documentation/devicetree/bindings/i2c/i2c-aspeed.txt b/Documentation/devicetree/bindings/i2c/i2c-aspeed.txt
index 8fbd8633a387..42ecaaf67172 100644
--- a/Documentation/devicetree/bindings/i2c/i2c-aspeed.txt
+++ b/Documentation/devicetree/bindings/i2c/i2c-aspeed.txt
@@ -13,9 +13,13 @@ Required Properties:
 - interrupts		: interrupt number
 
 Optional Properties:
-- bus-frequency	: frequency of the bus clock in Hz defaults to 100 kHz when not
-		  specified
-- multi-master	: states that there is another master active on this bus.
+- bus-frequency		: frequency of the bus clock in Hz defaults to 100 kHz
+			  when not specified
+- multi-master		: states that there is another master active on this
+			  bus.
+- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
+			  multi-master is set, defaults to 100 ms when not
+			  specified.
 
 Example:
 
-- 
2.18.0

[RFC PATCH i2c-next 2/2] i2c: aspeed: Add bus idle waiting logic for multi-master use cases

[Jae Hyun Yoo]

In multi-master environment, this driver side master cannot know
exactly when peer master sends data to this side master so a case
can be happened that this master tries to send data through the
master_xfer function but slave data from peer master is still
being processed by this driver.

To prevent state corruption in the case, this patch adds checking
if any slave operation is ongoing and it wait up to the timeout
duration before starting a master_xfer operation.

Signed-off-by: Jae Hyun Yoo <jae.hyun.yoo@linux.intel.com>
---
 drivers/i2c/busses/i2c-aspeed.c | 70 ++++++++++++++++++++++++++-------
 1 file changed, 55 insertions(+), 15 deletions(-)

diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c
index c258c4d9a4c0..73359eda98be 100644
--- a/drivers/i2c/busses/i2c-aspeed.c
+++ b/drivers/i2c/busses/i2c-aspeed.c
@@ -12,6 +12,7 @@
 
 #include <linux/clk.h>
 #include <linux/completion.h>
+#include <linux/delay.h>
 #include <linux/err.h>
 #include <linux/errno.h>
 #include <linux/i2c.h>
@@ -99,6 +100,7 @@
 		 ASPEED_I2CD_INTR_TX_ACK)
 
 /* 0x14 : I2CD Command/Status Register   */
+#define ASPEED_I2CD_XFER_MODE_STS_MASK			GENMASK(22, 19)
 #define ASPEED_I2CD_SCL_LINE_STS			BIT(18)
 #define ASPEED_I2CD_SDA_LINE_STS			BIT(17)
 #define ASPEED_I2CD_BUS_BUSY_STS			BIT(16)
@@ -115,6 +117,10 @@
 /* 0x18 : I2CD Slave Device Address Register   */
 #define ASPEED_I2CD_DEV_ADDR_MASK			GENMASK(6, 0)
 
+/* Timeout for bus busy checking */
+#define ASPEED_I2CD_IDLE_WAIT_TIMEOUT_MS_DEFAULT	100   /* 100 ms */
+#define ASPEED_I2CD_BUS_BUSY_CHECK_INTERVAL_US		10000 /* 10 ms */
+
 enum aspeed_i2c_master_state {
 	ASPEED_I2C_MASTER_INACTIVE,
 	ASPEED_I2C_MASTER_START,
@@ -155,6 +161,9 @@ struct aspeed_i2c_bus {
 	int				cmd_err;
 	/* Protected only by i2c_lock_bus */
 	int				master_xfer_result;
+	/* Multi-master */
+	bool				multi_master;
+	u32				idle_wait_timeout_ms;
 #if IS_ENABLED(CONFIG_I2C_SLAVE)
 	struct i2c_client		*slave;
 	enum aspeed_i2c_slave_state	slave_state;
@@ -590,27 +599,49 @@ static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id)
 	return irq_remaining ? IRQ_NONE : IRQ_HANDLED;
 }
 
+static int aspeed_i2c_check_bus_busy(struct aspeed_i2c_bus *bus)
+{
+	u32 status_check_mask = ASPEED_I2CD_BUS_BUSY_STS;
+	ktime_t timeout;
+
+	if (bus->multi_master) {
+		might_sleep();
+		timeout = ktime_add_ms(ktime_get(), bus->idle_wait_timeout_ms);
+		/*
+		 * ASPEED_I2CD_XFER_MODE_STS_MASK is marked as
+		 * 'for debugging purpose only' in datasheet but ASPEED
+		 * confirmed that this reflects real information and good to be
+		 * used in practical code. It will be used only in multi-master
+		 * use cases.
+		 */
+		status_check_mask |= ASPEED_I2CD_XFER_MODE_STS_MASK;
+	}
+
+	for (;;) {
+		if (!(readl(bus->base + ASPEED_I2C_CMD_REG) &
+		      status_check_mask))
+			return 0;
+		if (!bus->multi_master)
+			break;
+		if (ktime_compare(ktime_get(), timeout) > 0)
+			break;
+		usleep_range((ASPEED_I2CD_BUS_BUSY_CHECK_INTERVAL_US >> 2) + 1,
+			     ASPEED_I2CD_BUS_BUSY_CHECK_INTERVAL_US);
+	}
+
+	return aspeed_i2c_recover_bus(bus);
+}
+
 static int aspeed_i2c_master_xfer(struct i2c_adapter *adap,
 				  struct i2c_msg *msgs, int num)
 {
 	struct aspeed_i2c_bus *bus = i2c_get_adapdata(adap);
 	unsigned long time_left, flags;
-	int ret = 0;
-
-	spin_lock_irqsave(&bus->lock, flags);
-	bus->cmd_err = 0;
 
-	/* If bus is busy, attempt recovery. We assume a single master
-	 * environment.
-	 */
-	if (readl(bus->base + ASPEED_I2C_CMD_REG) & ASPEED_I2CD_BUS_BUSY_STS) {
-		spin_unlock_irqrestore(&bus->lock, flags);
-		ret = aspeed_i2c_recover_bus(bus);
-		if (ret)
-			return ret;
-		spin_lock_irqsave(&bus->lock, flags);
-	}
+	if (aspeed_i2c_check_bus_busy(bus))
+		return -EAGAIN;
 
+	spin_lock_irqsave(&bus->lock, flags);
 	bus->cmd_err = 0;
 	bus->msgs = msgs;
 	bus->msgs_index = 0;
@@ -798,8 +829,17 @@ static int aspeed_i2c_init(struct aspeed_i2c_bus *bus,
 	if (ret < 0)
 		return ret;
 
-	if (!of_property_read_bool(pdev->dev.of_node, "multi-master"))
+	if (of_property_read_bool(pdev->dev.of_node, "multi-master")) {
+		bus->multi_master = true;
+		ret = of_property_read_u32(pdev->dev.of_node,
+					   "idle-wait-timeout-ms",
+					   &bus->idle_wait_timeout_ms);
+		if (ret)
+			bus->idle_wait_timeout_ms =
+				ASPEED_I2CD_IDLE_WAIT_TIMEOUT_MS_DEFAULT;
+	} else {
 		fun_ctrl_reg |= ASPEED_I2CD_MULTI_MASTER_DIS;
+	}
 
 	/* Enable Master Mode */
 	writel(readl(bus->base + ASPEED_I2C_FUN_CTRL_REG) | fun_ctrl_reg,
-- 
2.18.0

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
> +- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
> +			  multi-master is set, defaults to 100 ms when not
> +			  specified.

Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
property.

Jae

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 601 bytes --]

On Tue, Sep 18, 2018 at 11:02:54AM -0700, Jae Hyun Yoo wrote:
> On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
> > +- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
> > +			  multi-master is set, defaults to 100 ms when not
> > +			  specified.
> 
> Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
> property.

No need. This binding is not a HW description, so not a DT property in
my book. I still don't understand: Your IP core in master mode does not
have a BUSY bit or similar which detects when a START was detected and
clears after a STOP?


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

Hi Wolfram,

On 9/24/2018 2:58 PM, Wolfram Sang wrote:
> On Tue, Sep 18, 2018 at 11:02:54AM -0700, Jae Hyun Yoo wrote:
>> On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
>>> +- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
>>> +			  multi-master is set, defaults to 100 ms when not
>>> +			  specified.
>>
>> Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
>> property.
> 
> No need. This binding is not a HW description, so not a DT property in
> my book. I still don't understand: Your IP core in master mode does not
> have a BUSY bit or similar which detects when a START was detected and
> clears after a STOP?
> 

Okay, I'll keep this property as it is then.

Aspeed IP has the BUSY bit on the I2CD14 register and we are already
using the bit in the current code for single-master use cases.
Due to the bit doesn't cover well multi-master use cases so we need to
improve the current busy checking logic using the 'Transfer Mode State
Machine' bit fields on the same register through this patch set.

Thanks for the review!

Jae

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 989 bytes --]

On Mon, Sep 24, 2018 at 03:15:46PM -0700, Jae Hyun Yoo wrote:
> Hi Wolfram,
> 
> On 9/24/2018 2:58 PM, Wolfram Sang wrote:
> > On Tue, Sep 18, 2018 at 11:02:54AM -0700, Jae Hyun Yoo wrote:
> > > On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
> > > > +- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
> > > > +			  multi-master is set, defaults to 100 ms when not
> > > > +			  specified.
> > > 
> > > Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
> > > property.
> > 
> > No need. This binding is not a HW description, so not a DT property in
> > my book. I still don't understand: Your IP core in master mode does not
> > have a BUSY bit or similar which detects when a START was detected and
> > clears after a STOP?
> > 
> 
> Okay, I'll keep this property as it is then.

Sorry for the misunderstanding. I don't think this a property, at all.
It doesn't describe the hardware, it is more of a configuration thing,
or?


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

On 9/25/2018 1:27 AM, Wolfram Sang wrote:
> On Mon, Sep 24, 2018 at 03:15:46PM -0700, Jae Hyun Yoo wrote:
>> Hi Wolfram,
>>
>> On 9/24/2018 2:58 PM, Wolfram Sang wrote:
>>> On Tue, Sep 18, 2018 at 11:02:54AM -0700, Jae Hyun Yoo wrote:
>>>> On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
>>>>> +- idle-wait-timeout-ms	: bus idle waiting timeout in milliseconds when
>>>>> +			  multi-master is set, defaults to 100 ms when not
>>>>> +			  specified.
>>>>
>>>> Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
>>>> property.
>>>
>>> No need. This binding is not a HW description, so not a DT property in
>>> my book. I still don't understand: Your IP core in master mode does not
>>> have a BUSY bit or similar which detects when a START was detected and
>>> clears after a STOP?
>>>
>>
>> Okay, I'll keep this property as it is then.
> 
> Sorry for the misunderstanding. I don't think this a property, at all.
> It doesn't describe the hardware, it is more of a configuration thing,
> or?
> 

You are right. It doesn't describe the hardware but it needs to be
configurable because it very depends on the peer master's behavior.
If peer master sends a long packet usually, it should have a long
timeout value since a slave receiving operation takes long time,
and it should be adjusted with an optimal value with taking some
experiments to make it not too long. Any suggestion?

Thanks,
Jae

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

On 9/25/2018 9:20 AM, Jae Hyun Yoo wrote:
> On 9/25/2018 1:27 AM, Wolfram Sang wrote:
>> On Mon, Sep 24, 2018 at 03:15:46PM -0700, Jae Hyun Yoo wrote:
>>> Hi Wolfram,
>>>
>>> On 9/24/2018 2:58 PM, Wolfram Sang wrote:
>>>> On Tue, Sep 18, 2018 at 11:02:54AM -0700, Jae Hyun Yoo wrote:
>>>>> On 9/10/2018 2:45 PM, Jae Hyun Yoo wrote:
>>>>>> +- idle-wait-timeout-ms    : bus idle waiting timeout in 
>>>>>> milliseconds when
>>>>>> +              multi-master is set, defaults to 100 ms when not
>>>>>> +              specified.
>>>>>
>>>>> Will change it to 'aspeed,idle-wait-timeout-ms' as it's a non standard
>>>>> property.
>>>>
>>>> No need. This binding is not a HW description, so not a DT property in
>>>> my book. I still don't understand: Your IP core in master mode does not
>>>> have a BUSY bit or similar which detects when a START was detected and
>>>> clears after a STOP?
>>>>
>>>
>>> Okay, I'll keep this property as it is then.
>>
>> Sorry for the misunderstanding. I don't think this a property, at all.
>> It doesn't describe the hardware, it is more of a configuration thing,
>> or?
>>
> 
> You are right. It doesn't describe the hardware but it needs to be
> configurable because it very depends on the peer master's behavior.
> If peer master sends a long packet usually, it should have a long
> timeout value since a slave receiving operation takes long time,
> and it should be adjusted with an optimal value with taking some
> experiments to make it not too long. Any suggestion?
> 

Should I use timeout in struct i2c_adapter instead just like i2c-mpc
does?

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 103 bytes --]


> Should I use timeout in struct i2c_adapter instead just like i2c-mpc
> does?

I would accept that.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [RFC PATCH i2c-next 1/2] dt-bindings: i2c: aspeed: Add 'idle-wait-timeout-ms' setting

[Jae Hyun Yoo]

On 9/26/2018 9:27 AM, Wolfram Sang wrote:
> 
>> Should I use timeout in struct i2c_adapter instead just like i2c-mpc
>> does?
> 
> I would accept that.
> 

Okay, I will submit a new patch set. Thanks!

Jae