On Thu, Feb 14, 2013 at 12:24:49PM +0100, Sascha Hauer wrote:
> On Thu, Feb 14, 2013 at 12:15:10PM +0200, Felipe Balbi wrote:
> > Hi,
> >
> > On Thu, Feb 14, 2013 at 11:07:22AM +0100, Sascha Hauer wrote:
> > > > > >> @@ -32,4 +35,37 @@ const char *usb_speed_string(enum usb_device_speed speed)
> > > > > >> }
> > > > > >> EXPORT_SYMBOL_GPL(usb_speed_string);
> > > > > >>
> > > > > >> +#ifdef CONFIG_OF
> > > > > >> +static const char *usb_dr_modes[] = {
> > > > > >> + [USB_DR_MODE_UNKNOWN] = "",
> > > > > >> + [USB_DR_MODE_HOST] = "host",
> > > > > >> + [USB_DR_MODE_PERIPHERAL] = "peripheral",
> > > > > >> + [USB_DR_MODE_OTG] = "otg",
> > > > > >> +};
> > > > > >
> > > > > > It turns out this is a problem, especially since this is generic usb
> > > > > > code: we have a chipidea controller (a patchset just arrived) that does
> > > > > > both host and peripheral, but not otg. And I'm told now that dwc3
> > > > > > controller can be synthesized like that too.
> > > >
> > > > I wonder if this part is really necessary. Usually you would read it
> > > > from HW's registers. For dwc3, it's quite recently that we allowed the
> > > > driver to be built with host-only, device-only or DRD functionality.
> > >
> > > We have quite some boards on which the ID pin is not wired up, so if a
> > > core is both host and device capable there is no way to detect the
> > > wanted mode if not given from the devicetree.
> >
> > right, that's fair. But that doesn't mean board can't work as both,
> > right. The IP is still the same, just the board is wired differently ;-)
>
> Yes, it is. Usually I run kernels with both host and device support enabled.
> Consider the IP is OTG capable, the chipidea driver will initialize
> both roles. Now if a board only supports one role and does not have an
> ID pin, how do I make sure the driver is in the correct mode? I need to
> specify it somehow. Otherwise I may end up on a host-only board with the
> driver sitting in device mode, or with a device-only board with the
> driver sitting in host mode.
right.
> > > > Maybe we can ignore dr_mode in host-only and device-only builds and only
> > > > look at it for DRD builds ?
> > >
> > > If something is or is not compiled in the kernel this doesn't mean the kernel
> > > is not started on boards with a different situation.
> >
> > who said kernel wouldn't start ? If you request a host-only build, you
> > need to force your IP into working as host, since that's all you have,
> > either that or you bail out on probe().
>
> Let me clarify, I don't want to use Kconfig to specify my boards
> capabilities. If a kernel is compiled for host mode only and the
> devicetree specifies a port is device-only, then yes, the driver
> should bail out on probe, maybe leaving a message that it found
> a device for which the suitable role is not compiled in.
yeah, this is why I said we should ignore dr_mode (or bail out) when
!OTG.
--
balbi
Attachment:
signature.asc
Description: Digital signature
