[U-Boot] [PATCH RFC 0/2] usb: host: Add a wrapper layer for mutiple host support

[Simon Glass]

Hi,

On Jun 25, 2014 12:08 AM, "Marek Vasut" <marex at denx.de> wrote:
>
> On Wednesday, June 25, 2014 at 07:11:42 AM, Vivek Gautam wrote:
> > Hi Marek,
> >
> > On Tue, Jun 24, 2014 at 7:56 PM, Marek Vasut <marex at denx.de> wrote:
> > > On Tuesday, June 24, 2014 at 04:10:20 PM, Vivek Gautam wrote:
> > >> Hi Marek,
> > >>
> > >> It's been very long since we had discussion for introducing the
wrapper
> > >> layer to enable using multiple usb controller types.
> > >> Its been unfortunate that i had been busy with other tasks, and
couldn't
> > >> look into this.
> > >> Now that i got sometime, i have prepared a simple RFC patch which
right
> > >> now supports APIs translation for submit_control_msg(),
> > >> submit_bulk_msg(), submit_int_msg(), and usb_lowlevel_init() as well
as
> > >> usb_lowlevel_stop(). This was the simplest approach that could
> > >> differentiate between controller types.
> > >>
> > >> I had thought of another approach too, wherein there's a 'list'
passed
> > >> by the usb core layer, which would be filled with
'host_controller_drv'
> > >> structure, that would contain information about the driver. And then
> > >> each host controller driver will register certain callbacks that can
be
> > >> called from the upper layers. If you say i will send an RFC for this
> > >> approach.
> > >
> > > I think this approach in this patchset will not play well with the
driver
> > > model. Instead, I'd love to see a mean to instantiate each *HCI
> > > controller and have a USB core which would track those instances. The
> > > USB core would then be able to call whatever generic ops on those
> > > instances as needed. Does that make sense please ?
> >
> > True, i understand your point here. I think the second approach i was
> > talking of, goes in this direction.
> > I think i could not put it well in words there.
> >
> > I will prepare an RFC patch for that, and post it as soon as its
> > ready, so that you can have
> > a look.
>
> Ah, this would be so very appreciated! Thank you!

Should we consider just going straight for driver model?

Regards,
Simon