cmd: bind/unbind by class index doesn't work with usb gadget

[Simon Glass]

Hi Zixun,

On Fri, 26 Jul 2024 at 07:24, Zixun LI <admin at hifiphile.com> wrote:
>
> Bind/unbind a gadget function driver to a USB gadget device doesn't work
> with class index, while using node path is ok. For example:
>
> U-Boot> dm tree
>
>  Class     Index  Probed  Driver                Name
> -----------------------------------------------------------
>  root          0  [ + ]   root_driver           root_driver
>  simple_bus    0  [ + ]   simple_bus            |-- ahb
>  simple_bus    1  [ + ]   simple_bus            |   |-- apb
>  ...
>  usb           0  [   ]   atmel_usba_udc        |   |   |-- gadget at f803c000
>
> U-Boot> bind usb 0 usb_ether
> Cannot find device 0 of class usb
>
> Both USB host uclass and USB gadget uclass are using the same name "usb"
> while they are under different class id of UCLASS_USB and
> UCLASS_USB_GADGET_GENERIC.
>
> bind_by_class_index() of cmd/bind.c uses uclass_get_by_name() to find
> class id, since both class have the same name it will get the id of
> UCLASS_USB instead since it is at the front of the uclass id list.
>
> Then uclass_find_device() will try to find the wrong device and fail.
>
> But fixing this issue simply by renaming gadget uclass name will break
> the DT alias feature DM_UC_FLAG_SEQ_ALIAS.
>
> Any idea how to fix it ? Like adding a name_seq_alias field to
> uclass_driver structure ?

It looks like test/py/tests/test_bind.py does not cover this case.

We don't have a check for duplicate uclass names but probably should.
Also sandbox builds USB gadget support but I don't think it is used.
If it were we could add a test for this case.

I think the UDC uclass should be renamed to "usb_gadget". That may end
up requiring changes elsewhere, but I'm not sure.

There are also weird things like omap2430_musb having two different uclass IDs!

Regards,
SImon