Skip to contentSkip to conversation

i2c-gpio: Add support for deferred probing

PATCH Applied Jean Delvare 3 participants 8 messages Feb 28, 2013 – Mar 27, 2013 5 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i2c-gpio: Add support for deferred probing

GPIOs may not be available immediately when i2c-gpio looks for them. Implement support for deferred probing so that probing can be attempted again later when GPIO pins are finally available.

Cc Haavard Skinnemoen <[email protected]>

---
A little more changes were needed than I initially thought, because we want to check the pins before we allocate memory. Otherwise we would have a possibly large number of memory allocation and freeing cycles until GPIO pins are finally available.

I wrote this quite some time ago already, but I do not have any system to test it. I would appreciate if one or more users of the i2c-gpio driver could give it a try and confirm it works as intended, or at least doesn't introduce any regression. Thanks.

 drivers/i2c/busses/i2c-gpio.c |   75 ++++++++++++++++++++++++++++--------------
 1 file changed, 50 insertions(+), 25 deletions(-)
drivers/i2c/busses/i2c-gpio.c +50 -25
@@ -85,23 +85,29 @@ static int i2c_gpio_getscl(void *data)
 	return gpio_get_value(pdata->scl_pin);
 }
 
-static int of_i2c_gpio_probe(struct device_node *np,
-			     struct i2c_gpio_platform_data *pdata)
+static int of_i2c_gpio_get_pins(struct device_node *np,
+				unsigned int *sda_pin, unsigned int *scl_pin)
 {
-	u32 reg;
-
 	if (of_gpio_count(np) < 2)
 		return -ENODEV;
 
-	pdata->sda_pin = of_get_gpio(np, 0);
-	pdata->scl_pin = of_get_gpio(np, 1);
+	*sda_pin = of_get_gpio(np, 0);
+	*scl_pin = of_get_gpio(np, 1);
 
-	if (!gpio_is_valid(pdata->sda_pin) || !gpio_is_valid(pdata->scl_pin)) {
+	if (!gpio_is_valid(*sda_pin) || !gpio_is_valid(*scl_pin)) {
 		pr_err("%s: invalid GPIO pins, sda=%d/scl=%d\n",
-		       np->full_name, pdata->sda_pin, pdata->scl_pin);
+		       np->full_name, *sda_pin, *scl_pin);
 		return -ENODEV;
 	}
 
+	return 0;
+}
+
+static void of_i2c_gpio_get_props(struct device_node *np,
+				  struct i2c_gpio_platform_data *pdata)
+{
+	u32 reg;
+
 	of_property_read_u32(np, "i2c-gpio,delay-us", &pdata->udelay);
 
 	if (!of_property_read_u32(np, "i2c-gpio,timeout-ms", &reg))
@@ -113,8 +119,6 @@ static int of_i2c_gpio_probe(struct devi
 		of_property_read_bool(np, "i2c-gpio,scl-open-drain");
 	pdata->scl_is_output_only =
 		of_property_read_bool(np, "i2c-gpio,scl-output-only");
-
-	return 0;
 }
 
 static int i2c_gpio_probe(struct platform_device *pdev)
@@ -123,31 +127,52 @@ static int i2c_gpio_probe(struct platfor
 	struct i2c_gpio_platform_data *pdata;
 	struct i2c_algo_bit_data *bit_data;
 	struct i2c_adapter *adap;
+	unsigned int sda_pin, scl_pin;
 	int ret;
 
-	priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
-	if (!priv)
-		return -ENOMEM;
-	adap = &priv->adap;
-	bit_data = &priv->bit_data;
-	pdata = &priv->pdata;
-
+	/* First get the GPIO pins; if it fails, we'll defer the probe. */
 	if (pdev->dev.of_node) {
-		ret = of_i2c_gpio_probe(pdev->dev.of_node, pdata);
+		ret = of_i2c_gpio_get_pins(pdev->dev.of_node,
+					   &sda_pin, &scl_pin);
 		if (ret)
 			return ret;
 	} else {
 		if (!pdev->dev.platform_data)
 			return -ENXIO;
-		memcpy(pdata, pdev->dev.platform_data, sizeof(*pdata));
+		pdata = pdev->dev.platform_data;
+		sda_pin = pdata->sda_pin;
+		scl_pin = pdata->scl_pin;
 	}
 
-	ret = gpio_request(pdata->sda_pin, "sda");
-	if (ret)
+	ret = gpio_request(sda_pin, "sda");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */
 		goto err_request_sda;
-	ret = gpio_request(pdata->scl_pin, "scl");
-	if (ret)
+	}
+	ret = gpio_request(scl_pin, "scl");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */
 		goto err_request_scl;
+	}
+
+	priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
+	if (!priv) {
+		ret = -ENOMEM;
+		goto err_add_bus;
+	}
+	adap = &priv->adap;
+	bit_data = &priv->bit_data;
+	pdata = &priv->pdata;
+
+	if (pdev->dev.of_node) {
+		pdata->sda_pin = sda_pin;
+		pdata->scl_pin = scl_pin;
+		of_i2c_gpio_get_props(pdev->dev.of_node, pdata);
+	} else {
+		memcpy(pdata, pdev->dev.platform_data, sizeof(*pdata));
+	}
 
 	if (pdata->sda_is_open_drain) {
 		gpio_direction_output(pdata->sda_pin, 1);
@@ -211,9 +236,9 @@ static int i2c_gpio_probe(struct platfor
 	return 0;
 
 err_add_bus:
-	gpio_free(pdata->scl_pin);
+	gpio_free(scl_pin);
 err_request_scl:
-	gpio_free(pdata->sda_pin);
+	gpio_free(sda_pin);
 err_request_sda:
 	return ret;
 }
Signature · 1 line
Jean Delvare

A message that is not in the archive (sent off-list or not yet received): [email protected]

Tested-by
#2 lore raw

Hi Jean,

Quoted 12 lines · Jean Delvare

GPIOs may not be available immediately when i2c-gpio looks for them. Implement support for deferred probing so that probing can be attempted again later when GPIO pins are finally available.

Cc Haavard Skinnemoen <[email protected]>

---
A little more changes were needed than I initially thought, because we want to check the pins before we allocate memory. Otherwise we would have a possibly large number of memory allocation and freeing cycles until GPIO pins are finally available.

Quoted 7 lines

I wrote this quite some time ago already, but I do not have any system to test it. I would appreciate if one or more users of the i2c-gpio driver could give it a try and confirm it works as intended, or at least doesn't introduce any regression. Thanks.

  drivers/i2c/busses/i2c-gpio.c |   75 ++++++++++++++++++++++++++++--------------
  1 file changed, 50 insertions(+), 25 deletions(-)

Test on at91sam9g20ek_2mmc board, booting with non-dt kernel and dt-kernel base on Linux-3.8. Both are OK.

Best Regards,
Bo Shen

Quoted 129 lines
drivers/i2c/busses/i2c-gpio.c +50 -25
@@ -85,23 +85,29 @@ static int i2c_gpio_getscl(void *data)
  	return gpio_get_value(pdata->scl_pin);
  }
 
-static int of_i2c_gpio_probe(struct device_node *np,
-			     struct i2c_gpio_platform_data *pdata)
+static int of_i2c_gpio_get_pins(struct device_node *np,
+				unsigned int *sda_pin, unsigned int *scl_pin)
  {
-	u32 reg;
-
  	if (of_gpio_count(np) < 2)
  		return -ENODEV;
 
-	pdata->sda_pin = of_get_gpio(np, 0);
-	pdata->scl_pin = of_get_gpio(np, 1);
+	*sda_pin = of_get_gpio(np, 0);
+	*scl_pin = of_get_gpio(np, 1);
 
-	if (!gpio_is_valid(pdata->sda_pin) || !gpio_is_valid(pdata->scl_pin)) {
+	if (!gpio_is_valid(*sda_pin) || !gpio_is_valid(*scl_pin)) {
  		pr_err("%s: invalid GPIO pins, sda=%d/scl=%d\n",
-		       np->full_name, pdata->sda_pin, pdata->scl_pin);
+		       np->full_name, *sda_pin, *scl_pin);
  		return -ENODEV;
  	}
 
+	return 0;
+}
+
+static void of_i2c_gpio_get_props(struct device_node *np,
+				  struct i2c_gpio_platform_data *pdata)
+{
+	u32 reg;
+
  	of_property_read_u32(np, "i2c-gpio,delay-us", &pdata->udelay);
 
  	if (!of_property_read_u32(np, "i2c-gpio,timeout-ms", &reg))
@@ -113,8 +119,6 @@ static int of_i2c_gpio_probe(struct devi
  		of_property_read_bool(np, "i2c-gpio,scl-open-drain");
  	pdata->scl_is_output_only =
  		of_property_read_bool(np, "i2c-gpio,scl-output-only");
-
-	return 0;
  }
 
  static int i2c_gpio_probe(struct platform_device *pdev)
@@ -123,31 +127,52 @@ static int i2c_gpio_probe(struct platfor
  	struct i2c_gpio_platform_data *pdata;
  	struct i2c_algo_bit_data *bit_data;
  	struct i2c_adapter *adap;
+	unsigned int sda_pin, scl_pin;
  	int ret;
 
-	priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
-	if (!priv)
-		return -ENOMEM;
-	adap = &priv->adap;
-	bit_data = &priv->bit_data;
-	pdata = &priv->pdata;
-
+	/* First get the GPIO pins; if it fails, we'll defer the probe. */
  	if (pdev->dev.of_node) {
-		ret = of_i2c_gpio_probe(pdev->dev.of_node, pdata);
+		ret = of_i2c_gpio_get_pins(pdev->dev.of_node,
+					   &sda_pin, &scl_pin);
  		if (ret)
  			return ret;
  	} else {
  		if (!pdev->dev.platform_data)
  			return -ENXIO;
-		memcpy(pdata, pdev->dev.platform_data, sizeof(*pdata));
+		pdata = pdev->dev.platform_data;
+		sda_pin = pdata->sda_pin;
+		scl_pin = pdata->scl_pin;
  	}
 
-	ret = gpio_request(pdata->sda_pin, "sda");
-	if (ret)
+	ret = gpio_request(sda_pin, "sda");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */
  		goto err_request_sda;
-	ret = gpio_request(pdata->scl_pin, "scl");
-	if (ret)
+	}
+	ret = gpio_request(scl_pin, "scl");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */
  		goto err_request_scl;
+	}
+
+	priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
+	if (!priv) {
+		ret = -ENOMEM;
+		goto err_add_bus;
+	}
+	adap = &priv->adap;
+	bit_data = &priv->bit_data;
+	pdata = &priv->pdata;
+
+	if (pdev->dev.of_node) {
+		pdata->sda_pin = sda_pin;
+		pdata->scl_pin = scl_pin;
+		of_i2c_gpio_get_props(pdev->dev.of_node, pdata);
+	} else {
+		memcpy(pdata, pdev->dev.platform_data, sizeof(*pdata));
+	}
 
  	if (pdata->sda_is_open_drain) {
  		gpio_direction_output(pdata->sda_pin, 1);
@@ -211,9 +236,9 @@ static int i2c_gpio_probe(struct platfor
  	return 0;
 
  err_add_bus:
-	gpio_free(pdata->scl_pin);
+	gpio_free(scl_pin);
  err_request_scl:
-	gpio_free(pdata->sda_pin);
+	gpio_free(sda_pin);
  err_request_sda:
  	return ret;
  }

A message that is not in the archive (sent off-list or not yet received): [email protected]

#4 lore raw
Quoted 20 lines · Bo Shen

Hi Jean,

Quoted 17 lines · Jean Delvare

GPIOs may not be available immediately when i2c-gpio looks for them. Implement support for deferred probing so that probing can be attempted again later when GPIO pins are finally available.

Cc Haavard Skinnemoen <[email protected]>

---
A little more changes were needed than I initially thought, because we want to check the pins before we allocate memory. Otherwise we would have a possibly large number of memory allocation and freeing cycles until GPIO pins are finally available.

I wrote this quite some time ago already, but I do not have any system to test it. I would appreciate if one or more users of the i2c-gpio driver could give it a try and confirm it works as intended, or at least doesn't introduce any regression. Thanks.

Quoted 7 lines
Quoted 2 lines
  drivers/i2c/busses/i2c-gpio.c |   75 ++++++++++++++++++++++++++++--------------
  1 file changed, 50 insertions(+), 25 deletions(-)

Test on at91sam9g20ek_2mmc board, booting with non-dt kernel and dt-kernel base on Linux-3.8. Both are OK.

Thanks Bo, this is very appreciated!

Signature · 1 line
Jean Delvare
#5 lore raw
In reply to a message not in the archive

Hi Jean,

Quoted 23 lines · Jean Delvare

GPIOs may not be available immediately when i2c-gpio looks for them. Implement support for deferred probing so that probing can be attempted again later when GPIO pins are finally available.

Cc Haavard Skinnemoen <[email protected]>

---
A little more changes were needed than I initially thought, because we want to check the pins before we allocate memory. Otherwise we would have a possibly large number of memory allocation and freeing cycles until GPIO pins are finally available.

I wrote this quite some time ago already, but I do not have any system to test it. I would appreciate if one or more users of the i2c-gpio driver could give it a try and confirm it works as intended, or at least doesn't introduce any regression. Thanks.

 drivers/i2c/busses/i2c-gpio.c |   75 ++++++++++++++++++++++++++++--------------
 1 file changed, 50 insertions(+), 25 deletions(-)

drivers/i2c/busses/i2c-gpio.c static int i2c_gpio_getscl(void *data) · line 89

Quoted 7 lines
drivers/i2c/busses/i2c-gpio.c +2 -2
@@ -85,23 +85,29 @@ static int i2c_gpio_getscl(void *data)
 	return gpio_get_value(pdata->scl_pin);
 }
 
-static int of_i2c_gpio_probe(struct device_node *np,
-			     struct i2c_gpio_platform_data *pdata)
+static int of_i2c_gpio_get_pins(struct device_node *np,
+				unsigned int *sda_pin, unsigned int *scl_pin)

GPIOs are expressed via int.

drivers/i2c/busses/i2c-gpio.c static int i2c_gpio_probe(struct platfor · line 150

Quoted 4 lines
drivers/i2c/busses/i2c-gpio.c +4 -0
+	ret = gpio_request(sda_pin, "sda");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */

Would gpio_request_array() make the code simpler?

Thanks,

Wolfram

A message that is not in the archive (sent off-list or not yet received): 20130322115621.GB24508-z923LK4zBo2bacvFa/[email protected]

#7 lore raw

Hi Wolfram,

Thanks for the review.

Quoted 27 lines · Wolfram Sang
Quoted 26 lines · Jean Delvare

GPIOs may not be available immediately when i2c-gpio looks for them. Implement support for deferred probing so that probing can be attempted again later when GPIO pins are finally available.

Cc Haavard Skinnemoen <[email protected]>

---
A little more changes were needed than I initially thought, because we want to check the pins before we allocate memory. Otherwise we would have a possibly large number of memory allocation and freeing cycles until GPIO pins are finally available.

I wrote this quite some time ago already, but I do not have any system to test it. I would appreciate if one or more users of the i2c-gpio driver could give it a try and confirm it works as intended, or at least doesn't introduce any regression. Thanks.

 drivers/i2c/busses/i2c-gpio.c |   75 ++++++++++++++++++++++++++++--------------
 1 file changed, 50 insertions(+), 25 deletions(-)
drivers/i2c/busses/i2c-gpio.c +0 -0
@@ -85,23 +85,29 @@ static int i2c_gpio_getscl(void *data)
 	return gpio_get_value(pdata->scl_pin);
 }
 
Quoted 6 lines

drivers/i2c/busses/i2c-gpio.c static int i2c_gpio_getscl(void *data) · line 89

Quoted 4 lines
drivers/i2c/busses/i2c-gpio.c +2 -2
-static int of_i2c_gpio_probe(struct device_node *np,
-			     struct i2c_gpio_platform_data *pdata)
+static int of_i2c_gpio_get_pins(struct device_node *np,
+				unsigned int *sda_pin, unsigned int *scl_pin)

GPIOs are expressed via int.

What do you mean? In <linux/gpio.h> I see:

struct gpio {
	unsigned	gpio;

(...)

static inline int gpio_get_value(unsigned int gpio)

{
	return __gpio_get_value(gpio);
}

And in <linux/i2c-gpio.h>:

struct i2c_gpio_platform_data {
	unsigned int    sda_pin;
	unsigned int    scl_pin;

(...)

So unsigned int seems to be the right type for gpios. The OF layer indeed uses int as the return type of of_get_gpio(), presumably to be able to report errors, but the original code did not handle errors from these calls so I assumed they couldn't fail and did not bother adding error handling.

If you still have a concern about the types used, please clarify and let me know what change you expect.

Quoted 6 lines

drivers/i2c/busses/i2c-gpio.c static int i2c_gpio_probe(struct platfor · line 150

Quoted 4 lines
drivers/i2c/busses/i2c-gpio.c +4 -0
+	ret = gpio_request(sda_pin, "sda");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */

Would gpio_request_array() make the code simpler?

I gave it a try, this indeed makes the code slightly simpler (-4 lines) but the resulting binary slightly larger (+40 bytes on x86-64.) I'd say it's not worth it?

Note that my patch doesn't introduce the gpio_request() calls, they were there before, so this decision is actually independent from my patch, and even if we decide to switch to using gpio_request_array(), I'd rather do it in a separate patch for clarity.

Thanks,

Signature · 1 line
Jean Delvare

A message that is not in the archive (sent off-list or not yet received): [email protected]

#9 lore raw
Quoted 12 lines

What do you mean? In <linux/gpio.h> I see:

struct gpio {
	unsigned	gpio;

(...)

static inline int gpio_get_value(unsigned int gpio)

{
	return __gpio_get_value(gpio);
}
Quoted 6 lines

And in <linux/i2c-gpio.h>:

struct i2c_gpio_platform_data {
	unsigned int    sda_pin;
	unsigned int    scl_pin;

(...)

I remembered this paragraph from Documentation/gpio.txt:

===

If you want to initialize a structure with an invalid GPIO number, use some negative number (perhaps "-EINVAL"); that will never be valid. To test if such number from such a structure could reference a GPIO, you may use this predicate:

        int gpio_is_valid(int number);

...

===

Confusingly, I know found that the chapter starts with

===

GPIOs are identified by unsigned integers in the range 0..MAX_INT. That reserves "negative" numbers for other purposes like marking signals as "not available on this board", or indicating faults.

===

Meh.

Quoted 2 lines

If you still have a concern about the types used, please clarify and let me know what change you expect.

Leave it. I think the fragile part is gpio_is_valid() but this is truly outside the scope of this patch.

Quoted 10 lines
Quoted 6 lines

drivers/i2c/busses/i2c-gpio.c static int i2c_gpio_probe(struct platfor · line 150

Quoted 4 lines
drivers/i2c/busses/i2c-gpio.c +4 -0
+	ret = gpio_request(sda_pin, "sda");
+	if (ret) {
+		if (ret == -EINVAL)
+			ret = -EPROBE_DEFER;	/* Try again later */

Would gpio_request_array() make the code simpler?

I gave it a try, this indeed makes the code slightly simpler (-4 lines) but the resulting binary slightly larger (+40 bytes on x86-64.) I'd say it's not worth it?

OK. Then leave it.

Quoted 4 lines

Note that my patch doesn't introduce the gpio_request() calls, they were there before, so this decision is actually independent from my patch, and even if we decide to switch to using gpio_request_array(), I'd rather do it in a separate patch for clarity.

I don't fully get it. Do you want to appl gpio_request() to this patch? Otherwise, I'd take it as is.

Thanks,

Wolfram

A message that is not in the archive (sent off-list or not yet received): 20130326110908.GC8553-z923LK4zBo2bacvFa/[email protected]

#11 lore raw

Hi Wolfram,

Quoted 6 lines · Wolfram Sang
Quoted 2 lines

If you still have a concern about the types used, please clarify and let me know what change you expect.

Leave it. I think the fragile part is gpio_is_valid() but this is truly outside the scope of this patch.
(...)

Quoted 7 lines
Quoted 4 lines

Note that my patch doesn't introduce the gpio_request() calls, they were there before, so this decision is actually independent from my patch, and even if we decide to switch to using gpio_request_array(), I'd rather do it in a separate patch for clarity.

I don't fully get it. Do you want to appl gpio_request() to this patch? Otherwise, I'd take it as is.

As I do not understand your question, I'd say you take my patch as is :)

Thanks,

Signature · 1 line
Jean Delvare