1======================== 2LED handling under Linux 3======================== 4 5In its simplest form, the LED class just allows control of LEDs from 6userspace. LEDs appear in /sys/class/leds/. The maximum brightness of the 7LED is defined in max_brightness file. The brightness file will set the brightness 8of the LED (taking a value 0-max_brightness). Most LEDs don't have hardware 9brightness support so will just be turned on for non-zero brightness settings. 10 11The class also introduces the optional concept of an LED trigger. A trigger 12is a kernel based source of led events. Triggers can either be simple or 13complex. A simple trigger isn't configurable and is designed to slot into 14existing subsystems with minimal additional code. Examples are the disk-activity, 15nand-disk and sharpsl-charge triggers. With led triggers disabled, the code 16optimises away. 17 18Complex triggers while available to all LEDs have LED specific 19parameters and work on a per LED basis. The timer trigger is an example. 20The timer trigger will periodically change the LED brightness between 21LED_OFF and the current brightness setting. The "on" and "off" time can 22be specified via /sys/class/leds/<device>/delay_{on,off} in milliseconds. 23You can change the brightness value of a LED independently of the timer 24trigger. However, if you set the brightness value to LED_OFF it will 25also disable the timer trigger. 26 27You can change triggers in a similar manner to the way an IO scheduler 28is chosen (via /sys/class/leds/<device>/trigger). Trigger specific 29parameters can appear in /sys/class/leds/<device> once a given trigger is 30selected. 31 32 33Design Philosophy 34================= 35 36The underlying design philosophy is simplicity. LEDs are simple devices 37and the aim is to keep a small amount of code giving as much functionality 38as possible. Please keep this in mind when suggesting enhancements. 39 40 41LED Device Naming 42================= 43 44Is currently of the form: 45 46 "devicename:color:function" 47 48- devicename: 49 it should refer to a unique identifier created by the kernel, 50 like e.g. phyN for network devices or inputN for input devices, rather 51 than to the hardware; the information related to the product and the bus 52 to which given device is hooked is available in sysfs and can be 53 retrieved using get_led_device_info.sh script from tools/leds; generally 54 this section is expected mostly for LEDs that are somehow associated with 55 other devices. 56 57- color: 58 one of LED_COLOR_ID_* definitions from the header 59 include/dt-bindings/leds/common.h. 60 61- function: 62 one of LED_FUNCTION_* definitions from the header 63 include/dt-bindings/leds/common.h. 64 65If required color or function is missing, please submit a patch 66to linux-leds@vger.kernel.org. 67 68It is possible that more than one LED with the same color and function will 69be required for given platform, differing only with an ordinal number. 70In this case it is preferable to just concatenate the predefined LED_FUNCTION_* 71name with required "-N" suffix in the driver. fwnode based drivers can use 72function-enumerator property for that and then the concatenation will be handled 73automatically by the LED core upon LED class device registration. 74 75LED subsystem has also a protection against name clash, that may occur 76when LED class device is created by a driver of hot-pluggable device and 77it doesn't provide unique devicename section. In this case numerical 78suffix (e.g. "_1", "_2", "_3" etc.) is added to the requested LED class 79device name. 80 81There might be still LED class drivers around using vendor or product name 82for devicename, but this approach is now deprecated as it doesn't convey 83any added value. Product information can be found in other places in sysfs 84(see tools/leds/get_led_device_info.sh). 85 86Examples of proper LED names: 87 88 - "red:disk" 89 - "white:flash" 90 - "red:indicator" 91 - "phy1:green:wlan" 92 - "phy3::wlan" 93 - ":kbd_backlight" 94 - "input5::kbd_backlight" 95 - "input3::numlock" 96 - "input3::scrolllock" 97 - "input3::capslock" 98 - "mmc1::status" 99 - "white:status" 100 101get_led_device_info.sh script can be used for verifying if the LED name 102meets the requirements pointed out here. It performs validation of the LED class 103devicename sections and gives hints on expected value for a section in case 104the validation fails for it. So far the script supports validation 105of associations between LEDs and following types of devices: 106 107 - input devices 108 - ieee80211 compliant USB devices 109 110The script is open to extensions. 111 112There have been calls for LED properties such as color to be exported as 113individual led class attributes. As a solution which doesn't incur as much 114overhead, I suggest these become part of the device name. The naming scheme 115above leaves scope for further attributes should they be needed. If sections 116of the name don't apply, just leave that section blank. 117 118 119Keyboard backlight control LED Device Naming 120============================================ 121 122For backlit keyboards with a single brightness / color settings a single 123(multicolor) LED class device should be used to allow userspace to change 124the backlight brightness (and if possible the color). This LED class device 125must use "kbd_backlight" for the function part of the LED class device name. 126IOW the name must end with ":kbd_backlight". 127 128For backlit keyboards with multiple control zones, one (multicolor) LED class 129device should be used per zone. These LED class devices' name must follow: 130 131 "<devicename>:<color>:kbd_zoned_backlight-<zone_name>" 132 133and <devicename> must be the same for all zones of the same keyboard. 134 135<zone_name> should be descriptive of which part of the keyboard backlight 136the zone covers and should be suitable for userspace to show to an end user 137in an UI for controlling the zones. 138 139Where possible <zone_name> should be a value already used by other 140zoned keyboards with a similar or identical zone layout, e.g.: 141 142<devicename>:<color>:kbd_zoned_backlight-right 143<devicename>:<color>:kbd_zoned_backlight-middle 144<devicename>:<color>:kbd_zoned_backlight-left 145<devicename>:<color>:kbd_zoned_backlight-corners 146<devicename>:<color>:kbd_zoned_backlight-wasd 147 148or: 149 150<devicename>:<color>:kbd_zoned_backlight-main 151<devicename>:<color>:kbd_zoned_backlight-cursor 152<devicename>:<color>:kbd_zoned_backlight-numpad 153<devicename>:<color>:kbd_zoned_backlight-corners 154<devicename>:<color>:kbd_zoned_backlight-wasd 155 156Note that this is intended for keyboards with a limited number of zones, 157keyboards with per key addressable backlighting must not use LED class devices 158since the sysfs API is not suitable for rapidly change multiple LEDs in one 159"commit" as is necessary to do animations / special effects on such keyboards. 160 161An exception to the rule that all zones must follow: 162 163 "<devicename>:<color>:kbd_zoned_backlight-<zone_name>" 164 165is made for the special case where there is a single big zone which controls 166the backlighting of almost all of the keyboard and there are some small areas 167with separate control, like just the 4 cursor keys, or the WASD keys. In this 168case the main zone should use 'kbd_backlight' for the function part of the name 169for compatibility with (older) userspace code which is not aware of 170the "kbd_zoned_backlight-<zone_name>" function naming scheme. 171 172While the smaller zones should use the new zoned naming scheme. Such a setup 173would result in e.g.: 174 175<devicename>:<color>:kbd_backlight 176<devicename>:<color>:kbd_zoned_backlight-wasd 177 178"kbd_zoned_backlight-<zone_name>" aware userspace should be aware of this 179exception and check for a main zone with a "kbd_backlight" function-name. 180 181 182Brightness setting API 183====================== 184 185LED subsystem core exposes following API for setting brightness: 186 187 - led_set_brightness: 188 it is guaranteed not to sleep, passing LED_OFF stops 189 blinking, 190 191 - led_set_brightness_sync: 192 for use cases when immediate effect is desired - 193 it can block the caller for the time required for accessing 194 device registers and can sleep, passing LED_OFF stops hardware 195 blinking, returns -EBUSY if software blink fallback is enabled. 196 197 198LED registration API 199==================== 200 201A driver wanting to register a LED classdev for use by other drivers / 202userspace needs to allocate and fill a led_classdev struct and then call 203`[devm_]led_classdev_register`. If the non devm version is used the driver 204must call led_classdev_unregister from its remove function before 205free-ing the led_classdev struct. 206 207If the driver can detect hardware initiated brightness changes and thus 208wants to have a brightness_hw_changed attribute then the LED_BRIGHT_HW_CHANGED 209flag must be set in flags before registering. Calling 210led_classdev_notify_brightness_hw_changed on a classdev not registered with 211the LED_BRIGHT_HW_CHANGED flag is a bug and will trigger a WARN_ON. 212 213Hardware accelerated blink of LEDs 214================================== 215 216Some LEDs can be programmed to blink without any CPU interaction. To 217support this feature, a LED driver can optionally implement the 218blink_set() function (see <linux/leds.h>). To set an LED to blinking, 219however, it is better to use the API function led_blink_set(), as it 220will check and implement software fallback if necessary. 221 222To turn off blinking, use the API function led_brightness_set() 223with brightness value LED_OFF, which should stop any software 224timers that may have been required for blinking. 225 226The blink_set() function should choose a user friendly blinking value 227if it is called with `*delay_on==0` && `*delay_off==0` parameters. In this 228case the driver should give back the chosen value through delay_on and 229delay_off parameters to the leds subsystem. 230 231Setting the brightness to zero with brightness_set() callback function 232should completely turn off the LED and cancel the previously programmed 233hardware blinking function, if any. 234 235Hardware driven LEDs 236==================== 237 238Some LEDs can be programmed to be driven by hardware. This is not 239limited to blink but also to turn off or on autonomously. 240To support this feature, a LED needs to implement various additional 241ops and needs to declare specific support for the supported triggers. 242 243With hw control we refer to the LED driven by hardware. 244 245A sysfs attribute `trigger_may_offload_to_hw` is provided for userspace 246to query supported triggers and their states. 247 248LED driver must define the following value to support hw control: 249 250 - hw_control_trigger: 251 unique trigger name supported by the LED in hw control 252 mode. 253 254LED driver must implement the following API to support hw control: 255 - hw_control_is_supported: 256 check if the flags passed by the supported trigger can 257 be parsed and activate hw control on the LED. 258 259 Return 0 if the passed flags mask is supported and 260 can be set with hw_control_set(). 261 262 If the passed flags mask is not supported -EOPNOTSUPP 263 must be returned, the LED trigger will use software 264 fallback in this case. 265 266 Return a negative error in case of any other error like 267 device not ready or timeouts. 268 269 - hw_control_set: 270 activate hw control. LED driver will use the provided 271 flags passed from the supported trigger, parse them to 272 a set of mode and setup the LED to be driven by hardware 273 following the requested modes. 274 275 Set LED_OFF via the brightness_set to deactivate hw control. 276 277 Return 0 on success, a negative error number on failing to 278 apply flags. 279 280 - hw_control_get: 281 get active modes from a LED already in hw control, parse 282 them and set in flags the current active flags for the 283 supported trigger. 284 285 Return 0 on success, a negative error number on failing 286 parsing the initial mode. 287 Error from this function is NOT FATAL as the device may 288 be in a not supported initial state by the attached LED 289 trigger. 290 291 - hw_control_get_device: 292 return the device associated with the LED driver in 293 hw control. A trigger might use this to match the 294 returned device from this function with a configured 295 device for the trigger as the source for blinking 296 events and correctly enable hw control. 297 (example a netdev trigger configured to blink for a 298 particular dev match the returned dev from get_device 299 to set hw control) 300 301 Returns a pointer to a struct device or NULL if nothing 302 is currently attached. 303 304LED trigger should implement the following API to indicate hw control: 305 - hw_offloaded: 306 return a boolean indicating if the trigger is currently 307 offloaded to hardware. 308 309 If a trigger doesn't implement this callback, the default 310 value will be true for private triggers and false for generic 311 ones. 312 313LED driver can activate additional modes by default to workaround the 314impossibility of supporting each different mode on the supported trigger. 315Examples are hardcoding the blink speed to a set interval, enable special 316feature like bypassing blink if some requirements are not met. 317 318A trigger should first check if the hw control API are supported by the LED 319driver and check if the trigger is supported to verify if hw control is possible, 320use hw_control_is_supported to check if the flags are supported and only at 321the end use hw_control_set to activate hw control. 322 323A trigger can use hw_control_get to check if a LED is already in hw control 324and init their flags. 325 326Alternatively, a private trigger can be implemented along with the LED driver if 327the LED's hardware control doesn't fit any generic trigger. To associate the 328private trigger with the LED classdev, their `trigger_type` must be the same. To 329declare that the private trigger provides hardware control for the associated 330LED classdev, set the `hw_control_trigger` string to the trigger's name. Since 331both the LED classdev and the private trigger are in the same LED driver, it's 332not necessary for them to coordinate via `hw_control_*` callbacks. 333 334When the LED is in hw control, no software blink is possible and doing so 335will effectively disable hw control. 336 337Hardware-initiated trigger transition 338===================================== 339 340Some hardware can autonomously activate/deactivate hardware control. After that, 341the LED hardware notifies the LED driver. 342 343If the driver can detect such transitions and thus wants to notify the LED core 344to update the current trigger then the `LED_TRIG_HW_CHANGED` flag must be set in 345flags before registering. To update the current trigger accordingly, call 346`led_trigger_notify_hw_control_changed` on the LED classdev. 347 348This capability is restricted to the LED device's private trigger. The private 349trigger must have been properly registered (see above) and named after 350`hw_control_trigger`. 351 352Only two transitions are defined: 353 354- "none" => private trigger: 355 This happens when the hardware autonomously activates hardware control 356 and when "none" (i.e., no trigger) is currently active. If the private 357 trigger is already active when the method is called, this is essentially 358 a no-op. 359 360 The activation sequence for the private trigger will be executed as 361 normal. 362 363 The LED driver and its private trigger must be able to handle the 364 activation sequence even if the hardware is currently in hardware 365 control. 366 367 If error occurs in the activation sequence, the LED Trigger core reverts 368 the effective trigger to "none". 369 370- private trigger => "none" 371 This happens when the hardware autonomously deactivates hardware control 372 and when the private trigger is currently active. If "none" (i.e., no 373 trigger) is active when the method is called, this is essentially a 374 no-op. 375 376 The deactivation sequence for the private trigger will be executed as 377 normal, except that the current LED brightness is retained. The reason 378 for keeping the brightness unchanged is that some hardware may choose a 379 specific brightness instead of simply turning off the LED after 380 autonomously deactivating hardware control. 381 382 The LED driver and its private trigger must be able to handle the 383 deactivation sequence even if the hardware is not currently in hardware 384 control. 385 386If the current trigger is neither the private trigger nor "none", no transition 387will be made. 388 389Known Issues 390============ 391 392The LED Trigger core cannot be a module as the simple trigger functions 393would cause nightmare dependency issues. I see this as a minor issue 394compared to the benefits the simple trigger functionality brings. The 395rest of the LED subsystem can be modular. 396