Re: [PATCH] usb: gadget: udc: fix the kernel NULL pointer in composite_setup

[Felipe Balbi]

Hi,

"He, Bo" <bo.he@xxxxxxxxx> writes:
> the patch is for fix the below kernel panic:
> BUG: unable to handle kernel NULL pointer dereference at 000000000000002a
> IP: [<ffffffff8170e19d>] composite_setup+0x3d/0x1830
> PGD 27525b067 PUD 27525a067 PMD 0
> Oops: 0002 [#1] PREEMPT SMP
> Call Trace:
>  [<ffffffff8168b902>] ? dwc3_trace+0x52/0x60
>  [<ffffffff810c504d>] ? get_parent_ip+0xd/0x50
>  [<ffffffff8171159c>] android_setup+0xbc/0x140
>  [<ffffffff810ec190>] ? irq_finalize_oneshot+0xe0/0xe0
>  [<ffffffff816917e7>] dwc3_ep0_delegate_req+0x37/0x50
>  [<ffffffff81692e69>] dwc3_ep0_interrupt+0xaf9/0xc10
>  [<ffffffff810c504d>] ? get_parent_ip+0xd/0x50
>  [<ffffffff810ec190>] ? irq_finalize_oneshot+0xe0/0xe0
>  [<ffffffff81690281>] dwc3_thread_interrupt+0x931/0xbf0
>  [<ffffffff810ec190>] ? irq_finalize_oneshot+0xe0/0xe0
>  [<ffffffff810ec1ae>] irq_thread_fn+0x1e/0x40
>  [<ffffffff810ec674>] irq_thread+0x134/0x1b0
>  [<ffffffff810ec260>] ? wake_threads_waitq+0x30/0x30
>  [<ffffffff810b802d>] kthread+0xed/0x110
>  [<ffffffff81a2fd6f>] ret_from_fork+0x3f/0x70
> RIP  [<ffffffff8170e19d>] composite_setup+0x3d/0x1830
>
> the root cause is dwc interrupt comes after usb_gadget_remove_driver.
> the fix is stop udc to have the dwc3 disable the interrupt, then release
> the resource in udc->driver->unbind.
> usb_gadget_udc_stop-->
> udc->gadget->ops->udc_stop(udc->gadget)-->
> dwc3_gadget_stop
>
> Signed-off-by: he, bo <bo.he@xxxxxxxxx>
> ---
>  drivers/usb/gadget/udc/core.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c
> index e6f04ee..67e9aa5 100644
> --- a/drivers/usb/gadget/udc/core.c
> +++ b/drivers/usb/gadget/udc/core.c
> @@ -1258,8 +1258,8 @@ static void usb_gadget_remove_driver(struct usb_udc *udc)
>  
>  	usb_gadget_disconnect(udc->gadget);
>  	udc->driver->disconnect(udc->gadget);
> -	udc->driver->unbind(udc->gadget);
>  	usb_gadget_udc_stop(udc);
> +	udc->driver->unbind(udc->gadget);

unbind must be called before udc_stop. This seems to be a bug *only* in
dwc3. I can't see how this would happen, actually. On dwc3_gadget_stop()
we mask dwc3's interrupts, so the handler should be executed anymore.

Can you tell me how to reproduce this? I could try this out tomorrow.

Which kernel are you using? I wonder if this is something caused by the
Android patches.

-- 
balbi