2016-04-12 13:29:30

by Peter Griffin

[permalink] [raw]
Subject: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks

>From I2C specification:

http://www.nxp.com/documents/user_manual/UM10204.pdf

Rev. 03 section 3.1.16 titled "Bus clear". When the i2c device held the
SDA line low, the master should send 9 clocks pulses to try to recover.

Signed-off-by: Frederic Pillon <[email protected]>
Signed-off-by: Peter Griffin <[email protected]>
---
drivers/i2c/busses/i2c-st.c | 56 ++++++++++++++++++++++++++++++++++++++++++++-
1 file changed, 55 insertions(+), 1 deletion(-)

diff --git a/drivers/i2c/busses/i2c-st.c b/drivers/i2c/busses/i2c-st.c
index 6ee7715..37b0b34 100644
--- a/drivers/i2c/busses/i2c-st.c
+++ b/drivers/i2c/busses/i2c-st.c
@@ -337,10 +337,51 @@ static void st_i2c_hw_config(struct st_i2c_dev *i2c_dev)
writel_relaxed(val, i2c_dev->base + SSC_NOISE_SUPP_WIDTH_DATAOUT);
}

+/*
+ * i2c bus recovery routines
+ * get_scl and set_scl must be defined to avoid the recover_bus field of
+ * i2c_bus_recovery_info to be overriden with NULL during the
+ * i2c_add_adapter call
+ */
+
+static int st_i2c_get_scl(struct i2c_adapter *adap)
+{
+ return 0;
+}
+
+static void st_i2c_set_scl(struct i2c_adapter *adap, int val)
+{
+}
+
+static int st_i2c_recover_bus(struct i2c_adapter *i2c_adap)
+{
+ struct st_i2c_dev *i2c_dev = i2c_get_adapdata(i2c_adap);
+ u32 ctl;
+
+ dev_err(i2c_dev->dev, "Trying to recover bus\n");
+
+ /* Disable interrupts */
+ writel_relaxed(0, i2c_dev->base + SSC_IEN);
+
+ st_i2c_hw_config(i2c_dev);
+
+ ctl = SSC_CTL_EN | SSC_CTL_MS | SSC_CTL_EN_RX_FIFO | SSC_CTL_EN_TX_FIFO;
+ st_i2c_set_bits(i2c_dev->base + SSC_CTL, ctl);
+
+ st_i2c_clr_bits(i2c_dev->base + SSC_I2C, SSC_I2C_I2CM);
+ usleep_range(8000, 10000);
+
+ writel_relaxed(0, i2c_dev->base + SSC_TBUF);
+ usleep_range(2000, 4000);
+ st_i2c_set_bits(i2c_dev->base + SSC_I2C, SSC_I2C_I2CM);
+
+ return 0;
+}
+
static int st_i2c_wait_free_bus(struct st_i2c_dev *i2c_dev)
{
u32 sta;
- int i;
+ int i, ret;

for (i = 0; i < 10; i++) {
sta = readl_relaxed(i2c_dev->base + SSC_STA);
@@ -352,6 +393,12 @@ static int st_i2c_wait_free_bus(struct st_i2c_dev *i2c_dev)

dev_err(i2c_dev->dev, "bus not free (status = 0x%08x)\n", sta);

+ ret = i2c_recover_bus(&i2c_dev->adap);
+ if (ret) {
+ dev_err(i2c_dev->dev, "Failed to recover the bus (%d)\n", ret);
+ return ret;
+ }
+
return -EBUSY;
}

@@ -744,6 +791,12 @@ static struct i2c_algorithm st_i2c_algo = {
.functionality = st_i2c_func,
};

+static struct i2c_bus_recovery_info st_i2c_recovery_info = {
+ .recover_bus = st_i2c_recover_bus,
+ .get_scl = st_i2c_get_scl,
+ .set_scl = st_i2c_set_scl,
+};
+
static int st_i2c_of_get_deglitch(struct device_node *np,
struct st_i2c_dev *i2c_dev)
{
@@ -826,6 +879,7 @@ static int st_i2c_probe(struct platform_device *pdev)
adap->timeout = 2 * HZ;
adap->retries = 0;
adap->algo = &st_i2c_algo;
+ adap->bus_recovery_info = &st_i2c_recovery_info;
adap->dev.parent = &pdev->dev;
adap->dev.of_node = pdev->dev.of_node;

--
1.9.1


2016-04-24 21:10:52

by Wolfram Sang

[permalink] [raw]
Subject: Re: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks

> +/*
> + * i2c bus recovery routines
> + * get_scl and set_scl must be defined to avoid the recover_bus field of
> + * i2c_bus_recovery_info to be overriden with NULL during the
> + * i2c_add_adapter call
> + */

Oh, that shouldn't be like this. Can you try this patch and remove the
empty functions please?

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index 4979728f7fb2de..604936955807e5 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c
@@ -1595,10 +1595,12 @@ static int i2c_register_adapter(struct i2c_adapter *adap)

bri->get_scl = get_scl_gpio_value;
bri->set_scl = set_scl_gpio_value;
- } else if (!bri->set_scl || !bri->get_scl) {
+ } else if (bri->recover_bus == i2c_generic_scl_recovery) {
/* Generic SCL recovery */
- dev_err(&adap->dev, "No {get|set}_gpio() found, not using recovery\n");
- adap->bus_recovery_info = NULL;
+ if (!bri->set_scl || !bri->get_scl) {
+ dev_err(&adap->dev, "No {get|set}_scl() found, not using recovery\n");
+ adap->bus_recovery_info = NULL;
+ }
}
}


> +static int st_i2c_recover_bus(struct i2c_adapter *i2c_adap)
> +{

Can you describe what the function does? It is not clear to me that it
generates 9 scl pulses.

> + struct st_i2c_dev *i2c_dev = i2c_get_adapdata(i2c_adap);
> + u32 ctl;
> +
> + dev_err(i2c_dev->dev, "Trying to recover bus\n");

This is not an error. Either dev_dbg or drop it, I'd say.

> +
> + /* Disable interrupts */
> + writel_relaxed(0, i2c_dev->base + SSC_IEN);
> +
> + st_i2c_hw_config(i2c_dev);
> +
> + ctl = SSC_CTL_EN | SSC_CTL_MS | SSC_CTL_EN_RX_FIFO | SSC_CTL_EN_TX_FIFO;
> + st_i2c_set_bits(i2c_dev->base + SSC_CTL, ctl);
> +
> + st_i2c_clr_bits(i2c_dev->base + SSC_I2C, SSC_I2C_I2CM);
> + usleep_range(8000, 10000);
> +
> + writel_relaxed(0, i2c_dev->base + SSC_TBUF);
> + usleep_range(2000, 4000);
> + st_i2c_set_bits(i2c_dev->base + SSC_I2C, SSC_I2C_I2CM);
> +
> + return 0;
> +}

Thanks,

Wolfram


Attachments:
(No filename) (1.90 kB)
signature.asc (819.00 B)
Download all attachments

2016-04-28 13:31:10

by Maxime Coquelin

[permalink] [raw]
Subject: Re: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks

Hi Wolfram,

On 04/24/2016 11:10 PM, Wolfram Sang wrote:
>> +/*
>> + * i2c bus recovery routines
>> + * get_scl and set_scl must be defined to avoid the recover_bus field of
>> + * i2c_bus_recovery_info to be overriden with NULL during the
>> + * i2c_add_adapter call
>> + */
> Oh, that shouldn't be like this. Can you try this patch and remove the
> empty functions please?
>
> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> index 4979728f7fb2de..604936955807e5 100644
> --- a/drivers/i2c/i2c-core.c
> +++ b/drivers/i2c/i2c-core.c
> @@ -1595,10 +1595,12 @@ static int i2c_register_adapter(struct i2c_adapter *adap)
>
> bri->get_scl = get_scl_gpio_value;
> bri->set_scl = set_scl_gpio_value;
> - } else if (!bri->set_scl || !bri->get_scl) {
> + } else if (bri->recover_bus == i2c_generic_scl_recovery) {
> /* Generic SCL recovery */
> - dev_err(&adap->dev, "No {get|set}_gpio() found, not using recovery\n");
> - adap->bus_recovery_info = NULL;
> + if (!bri->set_scl || !bri->get_scl) {
> + dev_err(&adap->dev, "No {get|set}_scl() found, not using recovery\n");
> + adap->bus_recovery_info = NULL;
> + }
> }
> }
>
>
>> +static int st_i2c_recover_bus(struct i2c_adapter *i2c_adap)
>> +{
> Can you describe what the function does? It is not clear to me that it
> generates 9 scl pulses.
I agree, it would need some comments.
This IP is dual-role, it can do either SPI or I2C.
The trick is to switch to SPI mode, 9 bits words and write a 0,
so that 9 clock pulses are generated.

This is easier to manage than switching to GPIO mode,
as we don't need to provide the gpio handles in DT, and no need to
put/get the pinctrl handle.

Regards,
Maxime

2016-04-28 14:57:37

by Wolfram Sang

[permalink] [raw]
Subject: Re: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks


> The trick is to switch to SPI mode, 9 bits words and write a 0,
> so that 9 clock pulses are generated.

Heh. As long as it works :) But as you said, it really needs a comment.


Attachments:
(No filename) (181.00 B)
signature.asc (819.00 B)
Download all attachments

2016-04-29 11:26:50

by Maxime Coquelin

[permalink] [raw]
Subject: Re: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks



On 04/28/2016 04:57 PM, Wolfram Sang wrote:
>> The trick is to switch to SPI mode, 9 bits words and write a 0,
>> so that 9 clock pulses are generated.
> Heh. As long as it works :) But as you said, it really needs a comment.
>
:)
I didn't faced the problem myself, but it looks good with an oscilloscope,
and a customer reported it to work.

Peter, will you resend with adding the explanations I provided as comment?

Thanks in advance,
Maxime

2016-04-29 14:03:57

by Peter Griffin

[permalink] [raw]
Subject: Re: [PATCH] i2c: st: Implement i2c_bus_recovery_info callbacks

Hi Maxime,

On Fri, 29 Apr 2016, Maxime Coquelin wrote:

>
>
> On 04/28/2016 04:57 PM, Wolfram Sang wrote:
> >>The trick is to switch to SPI mode, 9 bits words and write a 0,
> >>so that 9 clock pulses are generated.
> >Heh. As long as it works :) But as you said, it really needs a comment.
> >
> :)
> I didn't faced the problem myself, but it looks good with an oscilloscope,
> and a customer reported it to work.
>
> Peter, will you resend with adding the explanations I provided as comment?
>
Yes I'll send a v2 with your comment added.

Peter