Re: [BUG REPORT] USB UDC/gadget regression

[Alan Stern]

On Wed, Feb 01, 2023 at 09:20:22PM +0200, Aaro Koskinen wrote:
> Hi,
> 
> After commit fc274c1e9973 ("USB: gadget: Add a new bus for gadgets"),
> g_ether started to crash on modprobe when used with OMAP UDC.
> 
> It seems that because of drv->bus getting set to NULL in the middle of
> bus_add_driver() things go wrong.
> 
> The below simple change seems to fix it, but I'm not sure if this is
> the correct way? Many drivers seem to be doing the same in UDC start -
> are they really all broken?

Gosh, yes... it's pretty surprising.  None of the UDC drivers should 
change driver->driver.bus.  I don't know why they ever did this; you 
would have expected that field to be NULL anyway before the fc274c1e9973 
commit because it would never get set to anything.

> diff --git a/drivers/usb/gadget/udc/omap_udc.c b/drivers/usb/gadget/udc/omap_udc.c
> index 6e9314e95be3..47d83bbb09e3 100644
> --- a/drivers/usb/gadget/udc/omap_udc.c
> +++ b/drivers/usb/gadget/udc/omap_udc.c
> @@ -2062,7 +2062,6 @@ static int omap_udc_start(struct usb_gadget *g,
>  	udc->softconnect = 1;
>  
>  	/* hook up the driver */
> -	driver->driver.bus = NULL;
>  	udc->driver = driver;
>  	spin_unlock_irqrestore(&udc->lock, flags);

If you would like to submit a patch fixing all the UDC drivers that do 
this, it would be great.

Alan Stern