Hi,
The commit bc882f5d5c7b4d6ed5e927bf838863af43c786e7 introduce auto probe of LED in binding function but that cause issue on my board.
This first patch of this patchset activateis the LED on my board to explain the issue, the second patch revert this commit and the third one propose an other solution.
For me, this commit is a error because in README of doc/driver-model/ it is indicated:
The device's bind() method is permitted to perform simple actions, but should not scan the device tree node, not initialise hardware, nor set up structures or allocate memory. All of these tasks should be left for the probe() method.
And in the patch the LED driver is probed during the binding scan.
When I activated the LED in my ARM target with the patch "stm32mp1: add gpio led support", I have the sequence:
dm_init_and_scan() :
1/ dm_scan_fdt_node() => loop on all the node
2/ scan for LED node => probe of LED driver is forced by "default-state" detection LED1 - "red" => probe of father of "red" node A - led B - soc C - root => probe of node needed by GPIO 1 - pin-controller 2 - gpio@50002000 3 - rcc-clk@50000000 4 - rcc@50000000
=> probe forced by default state for other led LED2 - green LED3 - orange
=> probe of node needed by GPIO (other bank) 5 - gpio@50009000
3/ dm_extended_scan_fdt scan continue... scan node "fixed-clock" under "/clocks" clk-hse clk-hsi clk-lse clk-lsi clk-csi
4/ probe of all the used devices.... after dm_extended_scan_fdt()
So many driver are probed before the end of the scan binding loop !
And that cause issue in my board (boot failed) because the rcc-clk clock driver found the input frequency with these fixed-clock, which are binded only after the stm32mp1 clock driver probe.
For me probe in forbidden in binding function and thus uclass_get_device_tail() is not allowed in the commit bc882f5d5c7b4d6ed5e927bf838863af43c786e7 which need to be reverted.
In the third patch I proposed an other solution based on the existing solution in u-class regulator used to enable regulator with "boot on" property (see regulators_enable_boot_on function).
I think that adding a function is the better solution in the driver model to force probe for some node according binding information (after dm_init_and_scan call).
This new function should be called in board_init function for each board but it could be also added in init_sequence_r[] in future.
Changes in v4: - simplify the uclass impact after Patrick Bruenn review - remove #ifdef CONFIG_LED for sandbox
Changes in v3: - add motivation in revert commit - minor update after Simon review - include led.h to avoid compilation warning on stm32mp1 board
Changes in v2: - add sandbox impact and test update
Patrick Delaunay (5): stm32mp1: add gpio led support Revert "dm: led: auto probe() LEDs with "default-state"" dm: led: move default state support in led uclass stm32mp1: use new function led default state sandbox: led: use new function to configure default state
arch/arm/dts/stm32mp157c-ed1-u-boot.dtsi | 24 ++++++++++++++++++++++++ board/sandbox/sandbox.c | 9 +++++++++ board/st/stm32mp1/stm32mp1.c | 4 ++++ common/board_r.c | 3 ++- configs/stm32mp15_basic_defconfig | 2 ++ drivers/led/led-uclass.c | 30 ++++++++++++++++++++++++++++++ drivers/led/led_gpio.c | 17 ----------------- include/led.h | 9 +++++++++ test/dm/led.c | 3 +++ 9 files changed, 83 insertions(+), 18 deletions(-)