On Wed, May 8, 2013 at 4:01 PM, Vladimir Zapolskiy
<vladimir_zapolskiy@xxxxxxxxxx> wrote:
> Hi Peter,
>
>
> On 05/07/13 04:39, Peter Chen wrote:
>>
>> On Fri, May 3, 2013 at 6:13 AM, Vladimir Zapolskiy
>> <vladimir_zapolskiy@xxxxxxxxxx> wrote:
>>>
>>> This change removes calls of
>>> uvc_function_connect()/uvc_function_disconnect()
>>> functions from open()/close() syscalls.
>>>
>>> This is a bugfix in g_webcam module, in some test scenarios (e.g. UDC in
>>> suspend mode or OTG in host mode) straightforward control of D+ pull-up
>>> might lead to system's misbehaviour.
>>>
>>
>> When UDC is in suspend mode, pull-up dp is controller's behavior.
>> Can you give an explanation how the system will go to abnormal
>> when UDC in suspend mode or OTG in host mode?
>
>
> one of the examples:
>
> 1. boot a kernel with runtime pm and g_webcam support, UDC driver supports
> runtime PM, no USB cable connected to a board
> 2. UDC goes to suspend mode, because there is no USB connections
> 3. g_webcam registers itself, /dev/video0 node is created
> 4. udev forks v4l_id, which opens /dev/video0 to query V4L2 capabilities
> 5. on open() UVC gadget commands UDC to pull-up D+ line, there is
> no connected USB cable though
> 6. the system stalls/panics after an attempt to access mapped UDC
> registers while UDC controller is in suspend.
>
> A regular .pullup function from a UDC controller driver looks like:
>
> ---8<---
> static int omap_pullup(struct usb_gadget *gadget, int is_on)
> {
> struct omap_udc *udc;
> unsigned long flags;
>
> udc = container_of(gadget, struct omap_udc, gadget);
> spin_lock_irqsave(&udc->lock, flags);
> udc->softconnect = (is_on != 0);
> if (can_pullup(udc))
> pullup_enable(udc);
> else
> pullup_disable(udc);
> spin_unlock_irqrestore(&udc->lock, flags);
> return 0;
> }
> ---8<---
>
> can_pullup() is false, pullup_disable() attempts to read/write to
> controller's registers, which causes system crash.
>
> There is definitely a bug, and according to Laurent's explanation UDC
> drivers should be fixed, I'll check that version.
>
Yes, it is UDC's bug. for UDC, which support runtime-pm, it should
consider the exported APIs might be called when udc is in suspend mode.
--
BR,
Peter Chen
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
