[PATCH v11 4/9] usb: chipidea: udc: add pullup/pulldown dp at hw_device_state

[Peter Chen]

On Thu, Mar 07, 2013 at 11:50:54AM +0200, Felipe Balbi wrote:
> Hi,
> 
> On Thu, Mar 07, 2013 at 10:36:13AM +0800, Peter Chen wrote:
> > > > diff --git a/drivers/usb/chipidea/udc.c b/drivers/usb/chipidea/udc.c
> > > > index e82dae4..70f9f2d 100644
> > > > --- a/drivers/usb/chipidea/udc.c
> > > > +++ b/drivers/usb/chipidea/udc.c
> > > > @@ -91,8 +91,10 @@ static int hw_device_state(struct ci13xxx *ci, u32 dma)
> > > >  		/* interrupt, error, port change, reset, sleep/suspend */
> > > >  		hw_write(ci, OP_USBINTR, ~0,
> > > >  			     USBi_UI|USBi_UEI|USBi_PCI|USBi_URI|USBi_SLI);
> > > > +		hw_write(ci, OP_USBCMD, USBCMD_RS, USBCMD_RS);
> > > >  	} else {
> > > >  		hw_write(ci, OP_USBINTR, ~0, 0);
> > > > +		hw_write(ci, OP_USBCMD, USBCMD_RS, 0);
> > > 
> > > this patch doesn't make sense to me. What will happen is that you will
> > > be enabling pullups when vbus_session() gets called and this might not
> > > be what gadget driver wants.
> > > 
> > > You don't want to fiddle with that yourself since I'm changing the
> > > framework so that gadget driver will always request pullups to be
> > > enabled.
> > 
> > Hi Felipe,
> > 
> > Do you think pullup dp without vbus is a valid operation?
> 
> why not ? What I want to connect pullups first and only then issue SRP ?

I am not familiar with OTG, but it only stands for special case, right?

> 
> > Current udc core code makes that possible.
> 
> so ?

Without vbus, but pullup dp, it will cause more power.

> 
> > But I think current your udc core code (add pullup after loading
> > gadget) make benefit for udc driver who has not vbus operation.
> > 
> > For chipidea driver:
> > 
> > - If vbus is there before load gadget module, the pullup dp is done by
> > udc core code.
> > - If vbus is not there before load gadget module, the pullup will be
> > done when the vbus is there.
> 
> This isn't legal. If you want to make sure vbus is alive before
> connecting pullups, then do it generically. Modify udc-core.c to make
> those checks for you. Bypassing the framework is dangerous because
> whenever I wanna change something, I might miss your private details and
> end up causing regressions.

Let thing be generic is a good idea. Then, is it ok I post a patch let
udc manage pullup by itself (through a flag) just you did for uvc?
UDC core doesn't know VBUS, so the pullup can't be managed by udc core
totally.

Besides, I looked four udc drivers (fsl_udc_core.c, at91_udc.c, mv_udc_core.c
and bcm63xx_udc.c), the first three manage pullup by itself, ony
bcm doesn't control it by itself.

> -- 
> balbi



-- 

Best Regards,
Peter Chen