Skip to contentSkip to conversation

iio: tsl2772: Use device-managed API

PATCH Superseded Chuhong Yuan 4 participants 8 messages Jul 26, 2019 – Jul 29, 2019 IIO

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iio: tsl2772: Use device-managed API

Use devm_iio_device_register to simplify
the code.

---
 drivers/iio/light/tsl2772.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/iio/light/tsl2772.c +1 -3
index 83cece921843..aa5891d105e8 100644
@@ -1877,7 +1877,7 @@ static int tsl2772_probe(struct i2c_client *clientp,
 	if (ret < 0)
 		return ret;
 
-	ret = iio_device_register(indio_dev);
+	ret = devm_iio_device_register(&clientp->dev, indio_dev);
 	if (ret) {
 		tsl2772_chip_off(indio_dev);
 		dev_err(&clientp->dev,
@@ -1928,8 +1928,6 @@ static int tsl2772_remove(struct i2c_client *client)
 
 	tsl2772_chip_off(indio_dev);
 
-	iio_device_unregister(indio_dev);
-
 	return 0;
 }
 
Signature · 1 line
2.20.1
maintainer
#1 lore raw
Quoted 4 lines · On Fri, 26 Jul 2019 20:30:58 +0800 Chuhong Yuan <[email protected]> wrote:

Use devm_iio_device_register to simplify
the code.

Please try to pick up on likely reviewers in your cc list. In this case Brian did a lot of work cleaning these drivers up so I've added him. Not everyone keeps up with the linux-iio list for some reason ;)

Immediate thought was that you had broken the ordering but turns out it is already buggy.

As things stand, we remove the userspace interfaces (iio_device_unregister)
after turning off the power. This is obviously a bad idea and also
form a purely "obviously correct" stand point, we aren't doing the reverse of probe.

So, I 'think' the right option is to reorder remove so that the power left on until after the iio_device_unregister call. At that point, we can't use devm_iio_device_register as we'll have the same incorrect ordering we currently have.

Brian, you looked at this driver most recently. Thoughts?

Thanks,

Jonathan

Quoted 26 lines
---
 drivers/iio/light/tsl2772.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/iio/light/tsl2772.c +1 -3
index 83cece921843..aa5891d105e8 100644
@@ -1877,7 +1877,7 @@ static int tsl2772_probe(struct i2c_client *clientp,
 	if (ret < 0)
 		return ret;
 
-	ret = iio_device_register(indio_dev);
+	ret = devm_iio_device_register(&clientp->dev, indio_dev);
 	if (ret) {
 		tsl2772_chip_off(indio_dev);
 		dev_err(&clientp->dev,
@@ -1928,8 +1928,6 @@ static int tsl2772_remove(struct i2c_client *client)
 
 	tsl2772_chip_off(indio_dev);
 
-	iio_device_unregister(indio_dev);
-
 	return 0;
 }
 
#2 lore raw
Quoted 20 lines · Jonathan Cameron
Quoted 4 lines · On Fri, 26 Jul 2019 20:30:58 +0800 Chuhong Yuan <[email protected]> wrote:

Use devm_iio_device_register to simplify
the code.

Please try to pick up on likely reviewers in your cc list. In this case Brian did a lot of work cleaning these drivers up so I've added him. Not everyone keeps up with the linux-iio list for some reason ;)

Immediate thought was that you had broken the ordering but turns out it is already buggy.

As things stand, we remove the userspace interfaces (iio_device_unregister)
after turning off the power. This is obviously a bad idea and also
form a purely "obviously correct" stand point, we aren't doing the reverse of probe.

Quoted 6 lines

So, I 'think' the right option is to reorder remove so that the power left on until after the iio_device_unregister call. At that point, we can't use devm_iio_device_register as we'll have the same incorrect ordering we currently have.

Brian, you looked at this driver most recently. Thoughts?

devm_add_action() could be used in the probe function to schedule the call to tsl2772_chip_off(). That would eliminate the need for tsl2772_remove(). See tsl2772_disable_regulators_action() for an example in that driver.

Chuhong: Another potential cleanup to shrink the size of this driver is to move it over to the regulator_bulk_() API. I didn't realize that API existed at the time I introduced the regulator support. If you're interested in taking on that cleanup as well, I can test those changes for you if you don't have the hardware.

Brian

Quoted 26 lines
Quoted 26 lines
---
 drivers/iio/light/tsl2772.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/iio/light/tsl2772.c +1 -3
index 83cece921843..aa5891d105e8 100644
@@ -1877,7 +1877,7 @@ static int tsl2772_probe(struct i2c_client *clientp,
 	if (ret < 0)
 		return ret;
 
-	ret = iio_device_register(indio_dev);
+	ret = devm_iio_device_register(&clientp->dev, indio_dev);
 	if (ret) {
 		tsl2772_chip_off(indio_dev);
 		dev_err(&clientp->dev,
@@ -1928,8 +1928,6 @@ static int tsl2772_remove(struct i2c_client *client)
 
 	tsl2772_chip_off(indio_dev);
 
-	iio_device_unregister(indio_dev);
-
 	return 0;
 }
 
#3 lore raw
Quoted 27 lines · Brian Masney <[email protected]> 于2019年7月28日周日 下午4:31写道:
Quoted 25 lines · Jonathan Cameron
Quoted 4 lines · On Fri, 26 Jul 2019 20:30:58 +0800 Chuhong Yuan <[email protected]> wrote:

Use devm_iio_device_register to simplify
the code.

Please try to pick up on likely reviewers in your cc list. In this case Brian did a lot of work cleaning these drivers up so I've added him. Not everyone keeps up with the linux-iio list for some reason ;)

Immediate thought was that you had broken the ordering but turns out it is already buggy.

As things stand, we remove the userspace interfaces (iio_device_unregister)
after turning off the power. This is obviously a bad idea and also
form a purely "obviously correct" stand point, we aren't doing the reverse of probe.

So, I 'think' the right option is to reorder remove so that the power left on until after the iio_device_unregister call. At that point, we can't use devm_iio_device_register as we'll have the same incorrect ordering we currently have.

Quoted 7 lines
Quoted 1 line

Brian, you looked at this driver most recently. Thoughts?

devm_add_action() could be used in the probe function to schedule the call to tsl2772_chip_off(). That would eliminate the need for tsl2772_remove(). See tsl2772_disable_regulators_action() for an example in that driver.

I find that we can use devm_add_action_or_reset() for tsl2772_disable_regulators_action() to eliminate the fault handling code.

I am not sure whether devm_add_action() can be used for tsl2772_chip_off() because it returns an integer, not void. And the return value is used in several functions.

Quoted 8 lines

Chuhong: Another potential cleanup to shrink the size of this driver is to move it over to the regulator_bulk_() API. I didn't realize that API existed at the time I introduced the regulator support. If you're interested in taking on that cleanup as well, I can test those changes for you if you don't have the hardware.

Brian

Does that mean merging vdd_supply and vddio_supply to an array of regulator_bulk_data and utilize regulator_bulk_() API to operate them together?
If so, I can do that cleanup in next version.

I have an additional question that I find regulator_disable() is used in the end of many .remove functions of drivers, which hinders us to use devm functions.
One example is drivers/iio/light/gp2ap020a00f.c.
Is there any solution to this problem?

Regards,
Chuhong

Quoted 27 lines
Quoted 27 lines
Quoted 27 lines
---
 drivers/iio/light/tsl2772.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/iio/light/tsl2772.c +1 -3
index 83cece921843..aa5891d105e8 100644
@@ -1877,7 +1877,7 @@ static int tsl2772_probe(struct i2c_client *clientp,
    if (ret < 0)
            return ret;
 
-   ret = iio_device_register(indio_dev);
+   ret = devm_iio_device_register(&clientp->dev, indio_dev);
    if (ret) {
            tsl2772_chip_off(indio_dev);
            dev_err(&clientp->dev,
@@ -1928,8 +1928,6 @@ static int tsl2772_remove(struct i2c_client *client)
 
    tsl2772_chip_off(indio_dev);
 
-   iio_device_unregister(indio_dev);
-
    return 0;
 }
 
#4 lore raw
Quoted 7 lines · Chuhong Yuan
Quoted 5 lines · Brian Masney <[email protected]> 于2019年7月28日周日 下午4:31写道:

devm_add_action() could be used in the probe function to schedule the call to tsl2772_chip_off(). That would eliminate the need for tsl2772_remove(). See tsl2772_disable_regulators_action() for an example in that driver.

Quoted 6 lines

I find that we can use devm_add_action_or_reset() for tsl2772_disable_regulators_action() to eliminate the fault handling code.

I am not sure whether devm_add_action() can be used for tsl2772_chip_off() because it returns an integer, not void. And the return value is used in several functions.

I would add a wrapper function (tsl2772_chip_off_action?) with the expected declaration that calls tsl2772_chip_off().

Quoted 6 lines
Quoted 6 lines

Chuhong: Another potential cleanup to shrink the size of this driver is to move it over to the regulator_bulk_() API. I didn't realize that API existed at the time I introduced the regulator support. If you're interested in taking on that cleanup as well, I can test those changes for you if you don't have the hardware.

Quoted 6 lines
Quoted 2 lines

Brian

Does that mean merging vdd_supply and vddio_supply to an array of regulator_bulk_data and utilize regulator_bulk_() API to operate them together?

Yes.

Quoted 5 lines

I have an additional question that I find regulator_disable() is used in the end of many .remove functions of drivers, which hinders us to use devm functions.
One example is drivers/iio/light/gp2ap020a00f.c.
Is there any solution to this problem?

There are devm_regulator_*() variants of the regulator API available that you can use. Lots of other APIs in the kernel have devm variants to simply drivers.

Brian

#5 lore raw
Quoted 39 lines · On Mon, 29 Jul 2019 04:03:07 -0400 Brian Masney <[email protected]> wrote:
Quoted 7 lines · Chuhong Yuan
Quoted 5 lines · Brian Masney <[email protected]> 于2019年7月28日周日 下午4:31写道:

devm_add_action() could be used in the probe function to schedule the call to tsl2772_chip_off(). That would eliminate the need for tsl2772_remove(). See tsl2772_disable_regulators_action() for an example in that driver.

Quoted 6 lines

I find that we can use devm_add_action_or_reset() for tsl2772_disable_regulators_action() to eliminate the fault handling code.

I am not sure whether devm_add_action() can be used for tsl2772_chip_off() because it returns an integer, not void. And the return value is used in several functions.

I would add a wrapper function (tsl2772_chip_off_action?) with the expected declaration that calls tsl2772_chip_off().

Quoted 6 lines
Quoted 6 lines

Chuhong: Another potential cleanup to shrink the size of this driver is to move it over to the regulator_bulk_() API. I didn't realize that API existed at the time I introduced the regulator support. If you're interested in taking on that cleanup as well, I can test those changes for you if you don't have the hardware.

Quoted 6 lines
Quoted 2 lines

Brian

Does that mean merging vdd_supply and vddio_supply to an array of regulator_bulk_data and utilize regulator_bulk_() API to operate them together?

Yes.

Quoted 5 lines

I have an additional question that I find regulator_disable() is used in the end of many .remove functions of drivers, which hinders us to use devm functions.
One example is drivers/iio/light/gp2ap020a00f.c.
Is there any solution to this problem?

Quoted 3 lines

There are devm_regulator_*() variants of the regulator API available that you can use. Lots of other APIs in the kernel have devm variants to simply drivers.

I don't think there are any devm_ versions of regulator disable.

IIRC the argument made when this last came up was that it was rarely correct
to be as course grained as a lot of IIO drivers are. We should probably
do runtime pm and turn these regulators off a lot more frequently.

The reality is that it is an optimization that doesn't get done in IIO drivers that often as we mostly just want them to work and many usecases aren't actually power constrained,

So we end up doing a lot of devm_add_action_or_reset to power down the regulators.

Jonathan

Quoted 2 lines

Brian

#6 lore raw
Quoted 52 lines · Jonathan Cameron <[email protected]> 于2019年7月29日周一 下午7:09写道:
Quoted 39 lines · On Mon, 29 Jul 2019 04:03:07 -0400 Brian Masney <[email protected]> wrote:
Quoted 7 lines · Chuhong Yuan
Quoted 5 lines · Brian Masney <[email protected]> 于2019年7月28日周日 下午4:31写道:

devm_add_action() could be used in the probe function to schedule the call to tsl2772_chip_off(). That would eliminate the need for tsl2772_remove(). See tsl2772_disable_regulators_action() for an example in that driver.

Quoted 6 lines

I find that we can use devm_add_action_or_reset() for tsl2772_disable_regulators_action() to eliminate the fault handling code.

I am not sure whether devm_add_action() can be used for tsl2772_chip_off() because it returns an integer, not void. And the return value is used in several functions.

I would add a wrapper function (tsl2772_chip_off_action?) with the expected declaration that calls tsl2772_chip_off().

Quoted 6 lines
Quoted 6 lines

Chuhong: Another potential cleanup to shrink the size of this driver is to move it over to the regulator_bulk_() API. I didn't realize that API existed at the time I introduced the regulator support. If you're interested in taking on that cleanup as well, I can test those changes for you if you don't have the hardware.

Quoted 6 lines
Quoted 2 lines

Brian

Does that mean merging vdd_supply and vddio_supply to an array of regulator_bulk_data and utilize regulator_bulk_() API to operate them together?

Yes.

Quoted 5 lines

I have an additional question that I find regulator_disable() is used in the end of many .remove functions of drivers, which hinders us to use devm functions.
One example is drivers/iio/light/gp2ap020a00f.c.
Is there any solution to this problem?

Quoted 3 lines

There are devm_regulator_*() variants of the regulator API available that you can use. Lots of other APIs in the kernel have devm variants to simply drivers.

I don't think there are any devm_ versions of regulator disable.

IIRC the argument made when this last came up was that it was rarely correct
to be as course grained as a lot of IIO drivers are. We should probably
do runtime pm and turn these regulators off a lot more frequently.

Quoted 7 lines

The reality is that it is an optimization that doesn't get done in IIO drivers that often as we mostly just want them to work and many usecases aren't actually power constrained,

So we end up doing a lot of devm_add_action_or_reset to power down the regulators.

I think I can add devm_ versions of regulator_enable and disable.

Chuhong

Quoted 5 lines

Jonathan

Quoted 2 lines

Brian

#7 lore raw
In reply to Jonathan Cameron
Quoted 11 lines · Jonathan Cameron
Quoted 3 lines · On Mon, 29 Jul 2019 04:03:07 -0400 Brian Masney <[email protected]> wrote:

There are devm_regulator_*() variants of the regulator API available that you can use. Lots of other APIs in the kernel have devm variants to simply drivers.

I don't think there are any devm_ versions of regulator disable.

IIRC the argument made when this last came up was that it was rarely correct
to be as course grained as a lot of IIO drivers are. We should probably
do runtime pm and turn these regulators off a lot more frequently.

Quoted 6 lines

The reality is that it is an optimization that doesn't get done in IIO drivers that often as we mostly just want them to work and many usecases aren't actually power constrained,

So we end up doing a lot of devm_add_action_or_reset to power down the regulators.

That makes sense. I have an out-of-tree patch where I started to add runtime pm support to the tsl2772 driver around the time I was working on the staging cleanup. I was unsure of how to do this when the user configures an interrupt threshold via sysfs since we don't want to completely power off the chip in that case. At the time, I couldn't find any other examples in IIO that showed how to do that. I should dust off that patch and send it out as a RFC to get some feedback.

Brian