| Message ID | 20221122111124.6828-1-cniedermaier@dh-electronics.com |
|---|---|
| State | New |
| Headers | show |
| Series | dt-bindings: leds: Mark label property as deprecated | expand |
From: Marek Vasut [mailto:marex@denx.de] Sent: Friday, November 25, 2022 10:27 PM > On 11/22/22 13:23, Pavel Machek wrote: >> Hi! > > Hi, > Hi, >>> Mark the label property as deprecated as it is mentioned >>> in the description. >> >> Lets do it the other way around. Functions (etc) don't really provide >> good enough description of LED, and label is still needed. > > Can you please provide a clear explanation which property or approach is > the correct one for new DTs ? > > So far, the documentation states that "label" is deprecated, and users > should replace it with "function" and "color". I'm a little confused because the direction seems to have changed here. Do you want label and functions (etc.) to be able to be used equally side by side, or should one of them be preferred? I also do DTs and it would be good to know what to go for. Could you please provide further explanation? Regards Christoph
On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: > On 11/22/22 13:23, Pavel Machek wrote: > > Hi! > > Hi, > > > > Mark the label property as deprecated as it is mentioned > > > in the description. > > > > Lets do it the other way around. Functions (etc) don't really provide > > good enough description of LED, and label is still needed. > > Can you please provide a clear explanation which property or approach is the > correct one for new DTs ? > > So far, the documentation states that "label" is deprecated, and users > should replace it with "function" and "color". 'function' is what activity/operation the LED is associated with. It is a fixed set of strings which s/w may use. It is a replacement for 'linux,default-trigger'. 'label' is what is printed next to the LED for a human to read. 'label' can be anything and the OS shouldn't care what it is. They serve 2 different purposes. Rob
On Wed 2022-11-30 13:19:05, Rob Herring wrote: > On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: > > On 11/22/22 13:23, Pavel Machek wrote: > > > Hi! > > > > Hi, > > > > > > Mark the label property as deprecated as it is mentioned > > > > in the description. > > > > > > Lets do it the other way around. Functions (etc) don't really provide > > > good enough description of LED, and label is still needed. > > > > Can you please provide a clear explanation which property or approach is the > > correct one for new DTs ? > > > > So far, the documentation states that "label" is deprecated, and users > > should replace it with "function" and "color". > > 'function' is what activity/operation the LED is associated with. It is > a fixed set of strings which s/w may use. It is a replacement for > 'linux,default-trigger'. > > 'label' is what is printed next to the LED for a human to read. 'label' > can be anything and the OS shouldn't care what it is. Unfortunately, no. We use label as a path in /sys/class/leds. And it looks like integer "function" is not really adequate for describing what LED does. There are too many LEDs and not enough integers, and it is common to have same function ("activity") on multiple devices ("wifi", "mmc", "eth"). Best regards, Pavel
On Wed, Nov 30, 2022 at 1:27 PM Pavel Machek <pavel@ucw.cz> wrote: > > On Wed 2022-11-30 13:19:05, Rob Herring wrote: > > On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: > > > On 11/22/22 13:23, Pavel Machek wrote: > > > > Hi! > > > > > > Hi, > > > > > > > > Mark the label property as deprecated as it is mentioned > > > > > in the description. > > > > > > > > Lets do it the other way around. Functions (etc) don't really provide > > > > good enough description of LED, and label is still needed. > > > > > > Can you please provide a clear explanation which property or approach is the > > > correct one for new DTs ? > > > > > > So far, the documentation states that "label" is deprecated, and users > > > should replace it with "function" and "color". > > > > 'function' is what activity/operation the LED is associated with. It is > > a fixed set of strings which s/w may use. It is a replacement for > > 'linux,default-trigger'. > > > > 'label' is what is printed next to the LED for a human to read. 'label' > > can be anything and the OS shouldn't care what it is. > > Unfortunately, no. That's why I said 'shouldn't care', not 'doesn't care'. 'label' is also not just an LED property. It's used elsewhere, but unfortunately the LED subsystem makes more use of it than it perhaps should. > We use label as a path in /sys/class/leds. Yes, or node name if no label. That's still not really caring what the value of label is. At least the kernel doesn't. A well behaved userspace wouldn't either and doesn't for most classes. > And it looks like integer > "function" is not really adequate for describing what LED does. There > are too many LEDs and not enough integers, and it is common to have > same function ("activity") on multiple devices ("wifi", "mmc", "eth"). Whatever the problems are, 'label' is not the solution. There is a way to associate leds with devices. 'trigger-source' IIRC. Rob
On 11/30/22 20:19, Rob Herring wrote: > On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: >> On 11/22/22 13:23, Pavel Machek wrote: >>> Hi! >> >> Hi, >> >>>> Mark the label property as deprecated as it is mentioned >>>> in the description. >>> >>> Lets do it the other way around. Functions (etc) don't really provide >>> good enough description of LED, and label is still needed. >> >> Can you please provide a clear explanation which property or approach is the >> correct one for new DTs ? >> >> So far, the documentation states that "label" is deprecated, and users >> should replace it with "function" and "color". > > 'function' is what activity/operation the LED is associated with. It is > a fixed set of strings which s/w may use. It is a replacement for > 'linux,default-trigger'. Isn't this 'function' more of a standardized replacement for 'label' ? $ git grep LED_FUNCTION_ include/ ... include/dt-bindings/leds/common.h:#define LED_FUNCTION_PLAYER5 "player-5" include/dt-bindings/leds/common.h:#define LED_FUNCTION_ACTIVITY "activity" include/dt-bindings/leds/common.h:#define LED_FUNCTION_ALARM "alarm" include/dt-bindings/leds/common.h:#define LED_FUNCTION_BACKLIGHT "backlight" include/dt-bindings/leds/common.h:#define LED_FUNCTION_BLUETOOTH "bluetooth" include/dt-bindings/leds/common.h:#define LED_FUNCTION_BOOT "boot" ... Seems to me that ^ is closer to a "standardized" form of 'label' . The LED subsystem does not infer any behavior of those LEDs based on their 'function' property as far as I can tell, at least not in the way 'linux,default-trigger' behaves. > 'label' is what is printed next to the LED for a human to read. 'label' > can be anything and the OS shouldn't care what it is. This part I understand. What is not clear to me is, why is 'label' being un-deprecated. We newly have 'function', 'function-enumerator' and 'color' DT properties for LEDs, which seem to be standardized forms of describing what the LED is used for, which LED it is (if there are multiple), and color of that LED. This was previously described in the 'label' property, usually in free form of e.g. "beaglebone:green:usr2" . > They serve 2 different purposes. [...]
On 11/30/22 20:26, Pavel Machek wrote: > On Wed 2022-11-30 13:19:05, Rob Herring wrote: >> On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: >>> On 11/22/22 13:23, Pavel Machek wrote: >>>> Hi! >>> >>> Hi, >>> >>>>> Mark the label property as deprecated as it is mentioned >>>>> in the description. >>>> >>>> Lets do it the other way around. Functions (etc) don't really provide >>>> good enough description of LED, and label is still needed. >>> >>> Can you please provide a clear explanation which property or approach is the >>> correct one for new DTs ? >>> >>> So far, the documentation states that "label" is deprecated, and users >>> should replace it with "function" and "color". >> >> 'function' is what activity/operation the LED is associated with. It is >> a fixed set of strings which s/w may use. It is a replacement for >> 'linux,default-trigger'. >> >> 'label' is what is printed next to the LED for a human to read. 'label' >> can be anything and the OS shouldn't care what it is. > > Unfortunately, no. > > We use label as a path in /sys/class/leds. And it looks like integer > "function" is not really adequate for describing what LED does. There > are too many LEDs and not enough integers, and it is common to have > same function ("activity") on multiple devices ("wifi", "mmc", "eth"). The Documentation/devicetree/bindings/leds/common.yaml schema indicates that function is a string, not an integer: " 32 function: 33 description: 34 LED function. Use one of the LED_FUNCTION_* prefixed definitions 35 from the header include/dt-bindings/leds/common.h. If there is no 36 matching LED_FUNCTION available, add a new one. 37 $ref: /schemas/types.yaml#/definitions/string "
Hi! > > > > So far, the documentation states that "label" is deprecated, and users > > > > should replace it with "function" and "color". > > > > > > 'function' is what activity/operation the LED is associated with. It is > > > a fixed set of strings which s/w may use. It is a replacement for > > > 'linux,default-trigger'. > > > > > > 'label' is what is printed next to the LED for a human to read. 'label' > > > can be anything and the OS shouldn't care what it is. > > > > Unfortunately, no. > > That's why I said 'shouldn't care', not 'doesn't care'. > > 'label' is also not just an LED property. It's used elsewhere, but > unfortunately the LED subsystem makes more use of it than it perhaps > should. > > > We use label as a path in /sys/class/leds. > > Yes, or node name if no label. That's still not really caring what the > value of label is. At least the kernel doesn't. A well behaved > userspace wouldn't either and doesn't for most classes. A well behaved userspace needs that to tell what kind of LED it is. It is important to tell keyboard backlight from HDD activity LED and from fllash on back camera for example. > > And it looks like integer > > "function" is not really adequate for describing what LED does. There > > are too many LEDs and not enough integers, and it is common to have > > same function ("activity") on multiple devices ("wifi", "mmc", "eth"). > > Whatever the problems are, 'label' is not the solution. > > There is a way to associate leds with devices. 'trigger-source' > IIRC. Neither is trigger-source a solution. Can we have linux,sysfs-path or something? Pavel
Hi all, On 12/2/22 00:41, Marek Vasut wrote: > On 11/30/22 20:19, Rob Herring wrote: >> On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: >>> On 11/22/22 13:23, Pavel Machek wrote: >>>> Hi! >>> >>> Hi, >>> >>>>> Mark the label property as deprecated as it is mentioned >>>>> in the description. >>>> >>>> Lets do it the other way around. Functions (etc) don't really provide >>>> good enough description of LED, and label is still needed. >>> >>> Can you please provide a clear explanation which property or approach >>> is the >>> correct one for new DTs ? >>> >>> So far, the documentation states that "label" is deprecated, and users >>> should replace it with "function" and "color". >> >> 'function' is what activity/operation the LED is associated with. It is >> a fixed set of strings which s/w may use. It is a replacement for >> 'linux,default-trigger'. > > Isn't this 'function' more of a standardized replacement for 'label' ? Yes it is. Introduction of function and color properties aimed at standardizing LED naming. Before there was only 'label' used for that, with DT node name as fallback if 'label' property was not provided. With introduction of 'function' and 'color' label was deprecated in the sense that if the former two are present, they are used for composing the LED name. In LED documentation [0] people are encouraged to use definitions from include/dt-bindings/leds/common.h to keep LED naming uniform. It allows to avoid duplicates like "wlan" and "wifi". > $ git grep LED_FUNCTION_ include/ > ... > include/dt-bindings/leds/common.h:#define LED_FUNCTION_PLAYER5 "player-5" > include/dt-bindings/leds/common.h:#define LED_FUNCTION_ACTIVITY "activity" > include/dt-bindings/leds/common.h:#define LED_FUNCTION_ALARM "alarm" > include/dt-bindings/leds/common.h:#define LED_FUNCTION_BACKLIGHT > "backlight" > include/dt-bindings/leds/common.h:#define LED_FUNCTION_BLUETOOTH > "bluetooth" > include/dt-bindings/leds/common.h:#define LED_FUNCTION_BOOT "boot" > ... > > Seems to me that ^ is closer to a "standardized" form of 'label' . > > The LED subsystem does not infer any behavior of those LEDs based on > their 'function' property as far as I can tell, at least not in the way > 'linux,default-trigger' behaves. > >> 'label' is what is printed next to the LED for a human to read. 'label' >> can be anything and the OS shouldn't care what it is. > > This part I understand. What is not clear to me is, why is 'label' being > un-deprecated. It shouldn't be. It seems to be Pavel's ad-hoc decision. > We newly have 'function', 'function-enumerator' and 'color' DT > properties for LEDs, which seem to be standardized forms of describing > what the LED is used for, which LED it is (if there are multiple), and > color of that LED. This was previously described in the 'label' > property, usually in free form of e.g. "beaglebone:green:usr2" . > >> They serve 2 different purposes. > > [...] [0] Documentation/leds/leds-class.rst
From: Jacek Anaszewski [mailto:jacek.anaszewski@gmail.com] Sent: Monday, December 5, 2022 7:44 PM > > Hi all, > Hi all, > On 12/2/22 00:41, Marek Vasut wrote: >> On 11/30/22 20:19, Rob Herring wrote: >>> On Fri, Nov 25, 2022 at 10:26:30PM +0100, Marek Vasut wrote: >>>> On 11/22/22 13:23, Pavel Machek wrote: >>>>> Hi! >>>> >>>> Hi, >>>> >>>>>> Mark the label property as deprecated as it is mentioned >>>>>> in the description. >>>>> >>>>> Lets do it the other way around. Functions (etc) don't really provide >>>>> good enough description of LED, and label is still needed. >>>> >>>> Can you please provide a clear explanation which property or approach >>>> is the >>>> correct one for new DTs ? >>>> >>>> So far, the documentation states that "label" is deprecated, and users >>>> should replace it with "function" and "color". >>> >>> 'function' is what activity/operation the LED is associated with. It is >>> a fixed set of strings which s/w may use. It is a replacement for >>> 'linux,default-trigger'. >> >> Isn't this 'function' more of a standardized replacement for 'label' ? > > Yes it is. Introduction of function and color properties aimed at > standardizing LED naming. Before there was only 'label' used for that, > with DT node name as fallback if 'label' property was not provided. > With introduction of 'function' and 'color' label was deprecated in > the sense that if the former two are present, they are used for > composing the LED name. > > In LED documentation [0] people are encouraged to use definitions from > include/dt-bindings/leds/common.h to keep LED naming uniform. > It allows to avoid duplicates like "wlan" and "wifi". > >> $ git grep LED_FUNCTION_ include/ >> ... >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_PLAYER5 "player-5" >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_ACTIVITY "activity" >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_ALARM "alarm" >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_BACKLIGHT >> "backlight" >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_BLUETOOTH >> "bluetooth" >> include/dt-bindings/leds/common.h:#define LED_FUNCTION_BOOT "boot" >> ... >> >> Seems to me that ^ is closer to a "standardized" form of 'label' . >> >> The LED subsystem does not infer any behavior of those LEDs based on >> their 'function' property as far as I can tell, at least not in the way >> 'linux,default-trigger' behaves. >> >>> 'label' is what is printed next to the LED for a human to read. 'label' >>> can be anything and the OS shouldn't care what it is. >> >> This part I understand. What is not clear to me is, why is 'label' being >> un-deprecated. > > It shouldn't be. It seems to be Pavel's ad-hoc decision. Is there a majority agreement that the "label" property remains deprecated? If so, I would say we can mark the label as deprecated. On the other hand, the new generated standardized sysfs name does not seem to provide a full replacement for the "label" property. What is still missing? How can the current state be extended to find more acceptance? [...] Regards Christoph
Hi! > >> This part I understand. What is not clear to me is, why is 'label' being > >> un-deprecated. > > > > It shouldn't be. It seems to be Pavel's ad-hoc decision. > > Is there a majority agreement that the "label" property remains > deprecated? > If so, I would say we can mark the label as deprecated. > > On the other hand, the new generated standardized sysfs name does not seem > to provide a full replacement for the "label" property. > What is still missing? Having reasonable naming of the LEDs is pre-requisite for deprecating label property. And no, kernel community does not operate by "majority agreement"s. Best regards, Pavel
On 12/22/22 14:50, Pavel Machek wrote: > Hi! > >>>> This part I understand. What is not clear to me is, why is 'label' being >>>> un-deprecated. >>> >>> It shouldn't be. It seems to be Pavel's ad-hoc decision. >> >> Is there a majority agreement that the "label" property remains >> deprecated? > > >> If so, I would say we can mark the label as deprecated. >> >> On the other hand, the new generated standardized sysfs name does not seem >> to provide a full replacement for the "label" property. >> What is still missing? > > Having reasonable naming of the LEDs is pre-requisite for deprecating > label property. As far as I can tell, function and function-enumerator is the reasonable naming. Jacek seem to confirm that. I would say, label can be deprecated . What is the counter-argument for why it should NOT be deprecated ?
On Thu 2022-12-22 15:01:44, Marek Vasut wrote: > On 12/22/22 14:50, Pavel Machek wrote: > > Hi! > > > > > > > This part I understand. What is not clear to me is, why is 'label' being > > > > > un-deprecated. > > > > > > > > It shouldn't be. It seems to be Pavel's ad-hoc decision. > > > > > > Is there a majority agreement that the "label" property remains > > > deprecated? > > > > > > > If so, I would say we can mark the label as deprecated. > > > > > > On the other hand, the new generated standardized sysfs name does not seem > > > to provide a full replacement for the "label" property. > > > What is still missing? > > > > Having reasonable naming of the LEDs is pre-requisite for deprecating > > label property. > > As far as I can tell, function and function-enumerator is the reasonable > naming. Jacek seem to confirm that. I would say, label can be deprecated . > What is the counter-argument for why it should NOT be deprecated ? When the label is no longer neccessary for naming leds, it can be deprecated. AFAICT, that is currently not the case. Best regards, Pavel
On 12/22/22 15:56, Pavel Machek wrote: > On Thu 2022-12-22 15:01:44, Marek Vasut wrote: >> On 12/22/22 14:50, Pavel Machek wrote: >>> Hi! >>> >>>>>> This part I understand. What is not clear to me is, why is 'label' being >>>>>> un-deprecated. >>>>> >>>>> It shouldn't be. It seems to be Pavel's ad-hoc decision. >>>> >>>> Is there a majority agreement that the "label" property remains >>>> deprecated? >>> >>> >>>> If so, I would say we can mark the label as deprecated. >>>> >>>> On the other hand, the new generated standardized sysfs name does not seem >>>> to provide a full replacement for the "label" property. >>>> What is still missing? >>> >>> Having reasonable naming of the LEDs is pre-requisite for deprecating >>> label property. >> >> As far as I can tell, function and function-enumerator is the reasonable >> naming. Jacek seem to confirm that. I would say, label can be deprecated . >> What is the counter-argument for why it should NOT be deprecated ? > > When the label is no longer neccessary for naming leds, it can be > deprecated. AFAICT, that is currently not the case. I'm sorry, this is not a counter-argument, this is hand-waving . Do you have anything to back your claim that the label is currently still needed, contrary to what the DT bindings document claims for years? "This property is deprecated - use 'function' and 'color' properties instead. function-enumerator has no effect when this property is present." " commit c5d18dd6b64e09dd6984bda9bdd55160af537a8c Date: Sun Jun 9 20:19:04 2019 +0200 dt-bindings: leds: Add properties for LED name construction Introduce dedicated properties for conveying information about LED function and color. Mark old "label" property as deprecated. Additionally function-enumerator property is being provided for the cases when neither function nor color can be used for LED differentiation. " It seems the function and function-enumerator is very much the replacement for label, except standardized. If that's not the case, do elaborate. If there is a special case that is not covered by it, do point it out.
diff --git a/Documentation/devicetree/bindings/leds/common.yaml b/Documentation/devicetree/bindings/leds/common.yaml index f5c57a580078..c1ce846f7676 100644 --- a/Documentation/devicetree/bindings/leds/common.yaml +++ b/Documentation/devicetree/bindings/leds/common.yaml @@ -52,6 +52,7 @@ properties: $ref: /schemas/types.yaml#/definitions/uint32 label: + deprecated: true description: The label for this LED. If omitted, the label is taken from the node name (excluding the unit address). It has to uniquely identify a device, i.e.
Mark the label property as deprecated as it is mentioned in the description. Signed-off-by: Christoph Niedermaier <cniedermaier@dh-electronics.com> --- Cc: Pavel Machek <pavel@ucw.cz> Cc: Rob Herring <robh+dt@kernel.org> Cc: Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org> Cc: Jacek Anaszewski <jacek.anaszewski@gmail.com> Cc: Marek Vasut <marex@denx.de> Cc: kernel@dh-electronics.com Cc: linux-leds@vger.kernel.org Cc: devicetree@vger.kernel.org To: linux-kernel@vger.kernel.org --- Documentation/devicetree/bindings/leds/common.yaml | 1 + 1 file changed, 1 insertion(+)