pwm_bl: Add support for backlight enable regulator
Subsystems (3)
PATCH v4 1/1 pwm_bl: Add support for backlight enable regulator
The backlight enable regulator is specified in the device tree node for backlight.
---
Changed the name of the property from en-supply to enable-supply.
Made enable-supply a mandatory property. Changed the example in the bindings documentation accordingly.
Moved devm_regulator_get() to the probe function. Regulator is now requested unconditionally.
Changed IS_ERR_OR_NULL to IS_ERR per Alex's recommendation.
.../bindings/video/backlight/pwm-backlight.txt | 14 ++++++ drivers/video/backlight/pwm_bl.c | 52 ++++++++++++++++---- 2 files changed, 56 insertions(+), 10 deletions(-)
Documentation/devicetree/bindings/video/backlight/pwm-backlight.txt +14 -0
@@ -10,6 +10,11 @@ Required properties: last value in the array represents a 100% duty cycle (brightest). - default-brightness-level: the default brightness level (index into the array defined by the "brightness-levels" property) + - enable-supply: A phandle to the regulator device tree node. This + regulator will be turned on and off as the pwm is enabled and disabled. + Many backlights are enabled via a GPIO. In this case, we instantiate + a fixed regulator and give that to enable-supply. If a regulator + is not needed, then provide a dummy fixed regulator. Optional properties: - pwm-names: a list of names for the PWM devices specified in the
@@ -19,10 +24,19 @@ Optional properties: Example: + bl_en: fixed-regulator { + compatible = "regulator-fixed"; + regulator-name = "bl-en-supply"; + regulator-boot-on; + gpio = <&some_gpio>; + enable-active-high; + }; + backlight { compatible = "pwm-backlight"; pwms = <&pwm 0 5000000>; brightness-levels = <0 4 8 16 32 64 128 255>; default-brightness-level = <6>; + enable-supply = <&bl_en>; };
drivers/video/backlight/pwm_bl.c +42 -10
104 lines not on this page · raw message
Signature · 1 line
1.7.9.5
On Wed, Mar 06, 2013 at 09:17:18AM -0800, Andrew Chew wrote:
[...]
Quoted 5 lines
drivers/video/backlight/pwm_bl.c +0
-0
drivers/video/backlight/pwm_bl.c line 28
Quoted 9 lines
drivers/video/backlight/pwm_bl.c +2
-0@@ -20,10 +20,13 @@#include <linux/pwm.h> #include <linux/pwm_backlight.h> #include <linux/slab.h> +#include <linux/regulator/consumer.h> struct pwm_bl_data {struct pwm_device *pwm;struct device *dev; +struct regulator *en_supply;
Can this be renamed to enable_supply to match the DT property name?
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;
This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Quoted 17 lines
drivers/video/backlight/pwm_bl.c +17
-0+static void pwm_backlight_enable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev); + +pwm_enable (pb->pwm); + +if (pb->en_supply && !pb->en_supply_enabled) { +if (regulator_enable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to enable regulator" ); +else + pb->en_supply_enabled = true; + } +} + +static void pwm_backlight_disable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev);
drivers/video/backlight/pwm_bl.c struct pwm_bl_data { · line 67
Quoted 10 lines
drivers/video/backlight/pwm_bl.c +10
-0+ +if (pb->en_supply && pb->en_supply_enabled) { +if (regulator_disable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to disable regulator" ); +else + pb->en_supply_enabled = false; + } + +pwm_disable (pb->pwm); +}
Alex already brought this up: shouldn't the sequences be reversed. That is, when enabling the backlight, turn on the regulator first, then enable the PWM. When disabling, disable the PWM first, then turn off the regulator?
drivers/video/backlight/pwm_bl.c static int pwm_backlight_probe(struct platform_device *pdev) · line 246
Quoted 10 lines
drivers/video/backlight/pwm_bl.c +6
-0@@ -213,6 +238,13 @@ static int pwm_backlight_probe(struct platform_device *pdev) pb->exit = data->exit; pb->dev = &pdev->dev; + pb->en_supply =devm_regulator_get (&pdev->dev,"enable" ); +if (IS_ERR (pb->en_supply)) { + ret =PTR_ERR (pb->en_supply); + pb->en_supply =NULL ; + goto err_alloc; + }
This effectively makes the regulator mandatory, so the board files that use pwm-backlight need to be updated or otherwise will break.
Thierry
Quoted 5 lines · Thierry Reding
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Sorry for being obstinate - but I'm still not convinced we can get rid of it. I checked the regulator code, and as you mentioned in the previous version, calls to regulator_enable() and regulator_disable() *must* be balanced in this driver.
Without this variable we would call regulator_enable() every time pwm_backlight_enable() is called (and same thing when disabling). Now imagine the driver is asked to set the following intensities: 5, 12, then 0. You would have two calls to regulator_enable() but only one to regulator_disable(), which would result in the enable GPIO remaining active even though it would be shut down. Or I missed something obvious.
The regulator must be enabled/disabled on transitions from/to 0, and AFAICT there is no way for this driver to detect them.
Quoted 24 lines
Quoted 24 lines
drivers/video/backlight/pwm_bl.c +24
-0+static void pwm_backlight_enable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev); + +pwm_enable (pb->pwm); + +if (pb->en_supply && !pb->en_supply_enabled) { +if (regulator_enable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to enable regulator" ); +else + pb->en_supply_enabled = true; + } +} + +static void pwm_backlight_disable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev); + +if (pb->en_supply && pb->en_supply_enabled) { +if (regulator_disable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to disable regulator" ); +else + pb->en_supply_enabled = false; + }
Quoted 8 lines
drivers/video/backlight/pwm_bl.c struct pwm_bl_data { · line 67
Quoted 3 lines
drivers/video/backlight/pwm_bl.c +3
-0+ +pwm_disable (pb->pwm); +}Alex already brought this up: shouldn't the sequences be reversed. That is, when enabling the backlight, turn on the regulator first, then enable the PWM. When disabling, disable the PWM first, then turn off the regulator?
Actually the current sequence seems to make sense - the PWM is always active when the enable GPIO is switched. If we do the contrary, we might have a short time where the backlight is enabled without receiving anything from the PWM. Don't think that would be serious, but the current behavior is similar to e.g. panels which we enable only after a signal is available.
Quoted 4 lines
Quoted 4 lines
drivers/video/backlight/pwm_bl.c +0
-0@@ -213,6 +238,13 @@ static int pwm_backlight_probe(struct platform_device *pdev) pb->exit = data->exit; pb->dev = &pdev->dev;
Quoted 9 lines
drivers/video/backlight/pwm_bl.c static int pwm_backlight_probe(struct platform_device *pdev) · line 246
Quoted 6 lines
drivers/video/backlight/pwm_bl.c +6
-0+ pb->en_supply =devm_regulator_get (&pdev->dev,"enable" ); +if (IS_ERR (pb->en_supply)) { + ret =PTR_ERR (pb->en_supply); + pb->en_supply =NULL ; + goto err_alloc; + }This effectively makes the regulator mandatory, so the board files that use pwm-backlight need to be updated or otherwise will break.
Yes. Btw, should such changes go into the same patch? This seems difficult to split without breaking things at some point.
Alex.
Quoted 20 lines · Alex Courbot
Quoted 5 lines · Thierry Reding
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Sorry for being obstinate - but I'm still not convinced we can get rid of it. I checked the regulator code, and as you mentioned in the previous version, calls to
regulator_enable()andregulator_disable()*must* be balanced in this driver.Without this variable we would call
regulator_enable()every timepwm_backlight_enable()is called (and same thing when disabling). Now imagine the driver is asked to set the following intensities: 5, 12, then 0. You would have two calls toregulator_enable()but only one toregulator_disable(), which would result in the enable GPIO remaining active even though it would be shut down. Or I missed something obvious.
Quoted 2 lines
The regulator must be enabled/disabled on transitions from/to 0, and AFAICT there is no way for this driver to detect them.
Yes, that's true, but I don't think it should be solved for just this one regulator. Instead if we need to track the enable state we might as well track it for *any* resource so that the PWM isn't enabled/disabled twice either.
Quoted 33 lines
Quoted 24 lines
Quoted 24 lines
drivers/video/backlight/pwm_bl.c +24
-0+static void pwm_backlight_enable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev); + +pwm_enable (pb->pwm); + +if (pb->en_supply && !pb->en_supply_enabled) { +if (regulator_enable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to enable regulator" ); +else + pb->en_supply_enabled = true; + } +} + +static void pwm_backlight_disable (struct backlight_device *bl) +{ +struct pwm_bl_data *pb =dev_get_drvdata (&bl->dev); + +if (pb->en_supply && pb->en_supply_enabled) { +if (regulator_disable (pb->en_supply) !=0 ) +dev_warn (&bl->dev,"Failed to disable regulator" ); +else + pb->en_supply_enabled = false; + }Quoted 8 lines
drivers/video/backlight/pwm_bl.c struct pwm_bl_data { · line 67
Quoted 3 lines
drivers/video/backlight/pwm_bl.c +3
-0+ +pwm_disable (pb->pwm); +}Alex already brought this up: shouldn't the sequences be reversed. That is, when enabling the backlight, turn on the regulator first, then enable the PWM. When disabling, disable the PWM first, then turn off the regulator?
Quoted 6 lines
Actually the current sequence seems to make sense - the PWM is always active when the enable GPIO is switched. If we do the contrary, we might have a short time where the backlight is enabled without receiving anything from the PWM. Don't think that would be serious, but the current behavior is similar to e.g. panels which we enable only after a signal is available.
Okay, let's leave it like that then.
Quoted 14 lines
Quoted 4 lines
Quoted 4 lines
drivers/video/backlight/pwm_bl.c +0
-0@@ -213,6 +238,13 @@ static int pwm_backlight_probe(struct platform_device *pdev) pb->exit = data->exit; pb->dev = &pdev->dev;Quoted 9 lines
drivers/video/backlight/pwm_bl.c static int pwm_backlight_probe(struct platform_device *pdev) · line 246
Quoted 6 lines
drivers/video/backlight/pwm_bl.c +6
-0+ pb->en_supply =devm_regulator_get (&pdev->dev,"enable" ); +if (IS_ERR (pb->en_supply)) { + ret =PTR_ERR (pb->en_supply); + pb->en_supply =NULL ; + goto err_alloc; + }This effectively makes the regulator mandatory, so the board files that use pwm-backlight need to be updated or otherwise will break.
Quoted 2 lines
Yes. Btw, should such changes go into the same patch? This seems difficult to split without breaking things at some point.
I expect that if the changes are split up then the board-setup code changes need to be done prior to the driver change. Using the lookup tables should make this easy because they aren't tied to the platform data and can be added independently. The patches should probably go through the same subsystem tree to take care of the dependencies.
Keeping everything in one patch would work too, but it's certainly more chaotic.
Thierry
Quoted 30 lines
From: Thierry Reding [mailto:[email protected]]
Sent: Thursday, March 07, 2013 3:27 AM
To: Alex Courbot
Cc: Andrew Chew; [email protected]
Subject: Re: [PATCH 1/1 v4]pwm_bl: Add support for backlight enable regulator* PGP Signed by an unknown key
Quoted 20 lines · Alex Courbot
Quoted 5 lines · Thierry Reding
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Sorry for being obstinate - but I'm still not convinced we can get rid of it. I checked the regulator code, and as you mentioned in the previous version, calls to
regulator_enable()andregulator_disable()*must* be balanced in this driver.Without this variable we would call
regulator_enable()every timepwm_backlight_enable()is called (and same thing when disabling). Now imagine the driver is asked to set the following intensities: 5, 12, then 0. You would have two calls toregulator_enable()but only one toregulator_disable(), which would result in the enable GPIO remaining active even though it would be shut down. Or I missed something obvious.
Quoted 6 lines
Quoted 2 lines
The regulator must be enabled/disabled on transitions from/to 0, and AFAICT there is no way for this driver to detect them.
Yes, that's true, but I don't think it should be solved for just this one regulator. Instead if we need to track the enable state we might as well track it for *any* resource so that the PWM isn't enabled/disabled twice either.
That makes sense, but I'm confused due to previous comments. The most obvious way to do this seems to be to have a bool track the enable state. Do you still want me to do away with this bool? I can satisfy your very last comment by keeping the bool (renaming it to something more generic) and encapsulating the pwm_enable()/pwm_disable() call within. I'll send out v5 today to show what I mean.
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
This effectively makes the regulator mandatory, so the board files that use pwm-backlight need to be updated or otherwise will break.
Yes. Btw, should such changes go into the same patch? This seems difficult to split without breaking things at some point.
Quoted 8 lines
I expect that if the changes are split up then the board-setup code changes need to be done prior to the driver change. Using the lookup tables should make this easy because they aren't tied to the platform data and can be added independently. The patches should probably go through the same subsystem tree to take care of the dependencies.
Keeping everything in one patch would work too, but it's certainly more chaotic.
Am I supposed to handle those patches? I'm concerned that I don't have hardware to test properly, but I can give it a shot if it's my responsibility.
Quoted 37 lines · Andrew Chew
Quoted 30 lines
From: Thierry Reding [mailto:[email protected]]
Sent: Thursday, March 07, 2013 3:27 AM
To: Alex Courbot
Cc: Andrew Chew; [email protected]
Subject: Re: [PATCH 1/1 v4]pwm_bl: Add support for backlight enable regulator* PGP Signed by an unknown key
Quoted 20 lines · Alex Courbot
Quoted 5 lines · Thierry Reding
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Sorry for being obstinate - but I'm still not convinced we can get rid of it. I checked the regulator code, and as you mentioned in the previous version, calls to
regulator_enable()andregulator_disable()*must* be balanced in this driver.Without this variable we would call
regulator_enable()every timepwm_backlight_enable()is called (and same thing when disabling). Now imagine the driver is asked to set the following intensities: 5, 12, then 0. You would have two calls toregulator_enable()but only one toregulator_disable(), which would result in the enable GPIO remaining active even though it would be shut down. Or I missed something obvious.Quoted 6 lines
Quoted 2 lines
The regulator must be enabled/disabled on transitions from/to 0, and AFAICT there is no way for this driver to detect them.
Yes, that's true, but I don't think it should be solved for just this one regulator. Instead if we need to track the enable state we might as well track it for *any* resource so that the PWM isn't enabled/disabled twice either.
Quoted 5 lines
That makes sense, but I'm confused due to previous comments. The most obvious way to do this seems to be to have a bool track the enable state. Do you still want me to do away with this bool? I can satisfy your very last comment by keeping the bool (renaming it to something more generic) and encapsulating the pwm_enable()/pwm_disable() call within.
I think that's what Thierry meant, yes.
Quoted 11 lines
Quoted 8 lines
I expect that if the changes are split up then the board-setup code changes need to be done prior to the driver change. Using the lookup tables should make this easy because they aren't tied to the platform data and can be added independently. The patches should probably go through the same subsystem tree to take care of the dependencies.
Keeping everything in one patch would work too, but it's certainly more chaotic.
Am I supposed to handle those patches? I'm concerned that I don't have hardware to test properly, but I can give it a shot if it's my responsibility.
Yes, if you introduce incompatibilities you have the burden of performing the transition without breaking things at any single point of the git history. Since this is just about adding a dummy regulator, it should go fine even without testing. And in the event it does not, that's what linux-next is for.
Make sure you also update the dts of current device tree users, as they will break, too.
What I don't know is if you should update all users in one big patch, or instead provide one patch per platform changed. Maybe Thierry can provide some guidance here.
Alex.
Quoted 38 lines · Alex Courbot
Quoted 37 lines · Andrew Chew
Quoted 36 lines
From: Thierry Reding [mailto:[email protected]]
Sent: Thursday, March 07, 2013 3:27 AM
To: Alex Courbot
Cc: Andrew Chew; [email protected]
Subject: Re: [PATCH 1/1 v4]pwm_bl: Add support for backlight enable regulator* PGP Signed by an unknown key
Quoted 20 lines · Alex Courbot
Quoted 5 lines · Thierry Reding
drivers/video/backlight/pwm_bl.c line 29
Quoted 1 line
drivers/video/backlight/pwm_bl.c +1
-0+bool en_supply_enabled;This boolean can be dropped. As discussed in a previous thread, the pwm-backlight driver shouldn't need to know about any other uses of the regulator.
Sorry for being obstinate - but I'm still not convinced we can get rid of it. I checked the regulator code, and as you mentioned in the previous version, calls to
regulator_enable()andregulator_disable()*must* be balanced in this driver.Without this variable we would call
regulator_enable()every timepwm_backlight_enable()is called (and same thing when disabling). Now imagine the driver is asked to set the following intensities: 5, 12, then 0. You would have two calls toregulator_enable()but only one toregulator_disable(), which would result in the enable GPIO remaining active even though it would be shut down. Or I missed something obvious.Quoted 2 lines
The regulator must be enabled/disabled on transitions from/to 0, and AFAICT there is no way for this driver to detect them.
Yes, that's true, but I don't think it should be solved for just this one regulator. Instead if we need to track the enable state we might as well track it for *any* resource so that the PWM isn't enabled/disabled twice either.
Quoted 7 lines
Quoted 5 lines
That makes sense, but I'm confused due to previous comments. The most obvious way to do this seems to be to have a bool track the enable state. Do you still want me to do away with this bool? I can satisfy your very last comment by keeping the bool (renaming it to something more generic) and encapsulating the pwm_enable()/pwm_disable() call within.
I think that's what Thierry meant, yes.
Yes, it is. =)
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
I expect that if the changes are split up then the board-setup code changes need to be done prior to the driver change. Using the lookup tables should make this easy because they aren't tied to the platform data and can be added independently. The patches should probably go through the same subsystem tree to take care of the dependencies.
Keeping everything in one patch would work too, but it's certainly more chaotic.
Quoted 8 lines
Quoted 2 lines
Am I supposed to handle those patches? I'm concerned that I don't have hardware to test properly, but I can give it a shot if it's my responsibility.
Yes, if you introduce incompatibilities you have the burden of performing the transition without breaking things at any single point of the git history. Since this is just about adding a dummy regulator, it should go fine even without testing. And in the event it does not, that's what linux-next is for.
Right. We'll need an Acked-by from the board/machine maintainers anyway and if something still breaks we can always fix it after somebody's actually done the testing.
Quoted 6 lines
Make sure you also update the dts of current device tree users, as they will break, too.
What I don't know is if you should update all users in one big patch, or instead provide one patch per platform changed. Maybe Thierry can provide some guidance here.
I think it'd be good to split them up into per-architecture and per-machine. Per-board would probably be too much. That'll allow the respective maintainers to ack patches that touch their machines or boards without having them go through all other hunks too.
Thierry