[PATCH] usb: Make what it takes for "usbkbd" to work in "stdin" clear

[Dragan Simic]

On 2025-01-27 23:23, Marek Vasut wrote:
> On 1/27/25 11:20 PM, Dragan Simic wrote:
>> On 2025-01-27 22:55, Marek Vasut wrote:
>>> On 1/19/25 7:50 PM, Dragan Simic wrote:
>>>> Selecting the CONFIG_SYS_CONSOLE_ENV_OVERWRITE option actually 
>>>> prevents the
>>>> CONFIG_USB_KEYBOARD option from working as expected, i.e. USB 
>>>> keyboards end
>>>> up not working as console inputs when "usbkbd" is properly specified 
>>>> as part
>>>> of the value of the "stdin" environment variable.  Describe this 
>>>> clearly in
>>>> the two relevant Kconfig files, to prevent any possible confusion.
>>>> 
>>>> In more detail, the console_init_r() function ends up overwriting 
>>>> the "stdin"
>>>> environment variable so it contains only the devices available at 
>>>> that point
>>>> (which doesn't include "usbkbd"), while the "usb start" operation is 
>>>> performed
>>>> much later, at which point the probe_usb_keyboard() function cannot 
>>>> assign any
>>>> discovered USB keyboards to the console because "usbkbd" is no 
>>>> longer present
>>>> in the "stdin" environment variable.
>>> 
>>> Wouldn't it be better to patch console_init_r() , check if "stdin"
>>> variable contains 'usbkbd' and if so, start USB and register the
>>> keyboard as stdin device at this point ? This is called after
>>> relocation, so starting USB should be OK to do here. The behavior
>>> should then be less confusing to users too.
>> 
>> Please correct me if I'm wrong, but isn't "usb start" deliberately
>> left to be executed as desired by the supported boards, or as part
>> of distroboot, i.e. in include/config_distro_bootcmd.h?
> 
> It is, but if the user needs USB keyboard as input device, shouldn't
> we probe for that (which yes, will make the start up slower) ?

I agree that the way the whole thing works now, with or without this
patch, leaves a fair amount of room for confusion.  That's exactly
how I felt while writing this patch -- I was confused a bit about
the possible outcomes of different combinations of the configuration
options.  Though, I also felt that the way it works now also makes
the whole thing as flexible as possible, so the users can choose
whatever configuration they want, even the broken one.

I'm also afraid that, effectively, starting USB forcedly at that
point may cause some unforeseen issues or regressions.

It also crossed my mind that perhaps a good solution would be to
automatically deselect SYS_CONSOLE_ENV_OVERWRITE when the user
selects USB_KEYBOARD, because that's needed for USB keyboards to
actually work.  That should also have lower potential for causing
some issues or regressions.

Maybe another good option would be to modify the console_init_r()
function to simply leave "usbkbd" in the environment variables when
USB_KEYBOARD is selected, so the "usb start" executed later can
actually make USB keyboards work?  I think such an approach should
have the lowest potential for causing some issues or regressions,
while it would pretty much eliminate any confusion.