[PATCH 2/3 v4] arm: kirkwood: add dreamplug (fdt) support.
Jason
jason at lakedaemon.net
Fri Feb 24 03:12:10 EST 2012
On Thu, Feb 23, 2012 at 07:34:33AM +0000, Arnd Bergmann wrote:
> On Thursday 23 February 2012, Rob Herring wrote:
> > On 02/22/2012 01:18 PM, Jason Cooper wrote:
>
> > > diff --git a/arch/arm/mach-kirkwood/Kconfig b/arch/arm/mach-kirkwood/Kconfig
> > > index 7fc603b..6095884 100644
> > > --- a/arch/arm/mach-kirkwood/Kconfig
> > > +++ b/arch/arm/mach-kirkwood/Kconfig
> > > @@ -44,6 +44,20 @@ config MACH_GURUPLUG
> > > Say 'Y' here if you want your kernel to support the
> > > Marvell GuruPlug Reference Board.
> > >
> > > +config ARCH_KIRKWOOD_DT
> > > + bool "Marvell Kirkwood Flattened Device Tree"
> > > + select USE_OF
> > > + help
> > > + Say 'Y' here if you want your kernel to support the
> > > + Marvell Kirkwood using flattened device tree.
> > > +
> > > +config MACH_DREAMPLUG_DT
> > > + bool "Marvell DreamPlug (Flattened Device Tree)"
> > > + depends on ARCH_KIRKWOOD_DT
> > > + help
> > > + Say 'Y' here if you want your kernel to support the
> > > + Marvell DreamPlug (Flattened Device Tree).
> >
> > Why do you need 2 entries?
>
> The first one is used to build the board file for all DT based machines,
> the second one is used to select the dts file. I would be enough to have just
> the first one, but that makes it harder to find for people looking for
> dreamplug.
>
> Maybe change the text to "Generic Marvell Kirkwood DT based (e.g. DreamPlug)"?
I was thinking about this more, and had, I hope, a better idea: change
the 'depends on ARCH_KIRKWOOD_DT' to 'selects ARCH_KIRKWOOD_DT' and
remove the first entry. This way, the logic remains the same and the
word 'Dreamplug' is always visible. Any objections?
> > > +
> > > +static unsigned int dreamplug_mpp_config[] __initdata = {
> > > + MPP0_SPI_SCn,
> > > + MPP1_SPI_MOSI,
> > > + MPP2_SPI_SCK,
> > > + MPP3_SPI_MISO,
> > > + MPP47_GPIO, /* Bluetooth LED */
> > > + MPP48_GPIO, /* Wifi LED */
> > > + MPP49_GPIO, /* Wifi AP LED */
> > > + 0
> > > +};
> >
> > Do you need this to boot? All this data should come from the dtb.
>
> Putting this into the dtb would require converting kirkwood to use the
> pinctrl subsystem first, if we want to have proper bindings. When I
> did the hands-on review of the code with Jason during ELC, we decided
> to leave this being done the legacy way for now, which also matches
> how Tegra does it until we have the pinctrl bindings.
>
> Also note how the patch description says "Driver porting will start
> with the uart (see next patch), and progress from there. Possibly,
> spi/flash/partitions will be next." I think this is the best approach
> indeed and I'd probably leave the pinctrl stuff until the end.
Agreed.
> > > +
> > > +static void __init kirkwood_dt_init(void)
> > > +{
> > > + kirkwood_init();
> > > +
> > > + if (!of_machine_is_compatible("kirkwood,dreamplug"))
> >
> > Huh? Your string doesn't match your dts file.
> >
> > You've already matched against "marvell,dreamplug", so is this check
> > necessary?
>
> It's wrong, and the condition is negated. It should be
>
> if (of_machine_is_compatible("marvell,dreamplug"))
> dreamplug_init();
>
> The idea is that there is one init function for all dt based
> kirkwood machines, which contains special setup functions for
> those boards that don't (yet) describe all the hardware in the
> dt.
Good catch, will fix in v5.
> > > +
> > > +static const char *kirkwood_dt_board_compat[] = {
> > > + "marvell,dreamplug",
> > > + NULL
> > > +};
>
> Since you mention the name, it should probably be "globalscale,dreamplug"
> because the device is made by GlobalScale instead of Marvell.
hmm, Globalscale Tech is, as I understand it, a turn-key manufacturer.
They are simply building Marvell's development platforms for them,
handling sales, etc. My impression was that Marvell designed the board
and contracted Globalscale to build and distribute it.
I don't care which we use, but is the convention to use the SoC designer
(marvell,dreamplug), the SoC (kirkwood,dreamplug), or the brand
(globalscale,dreamplug)?
If there is no set standard, I think the SoC is most accurate, as
nothing prevents a manufacturer from swapping out ICs/SoCs between
manufacturing runs of the same make/model. Look at certain wifi USB
devices for examples.
thx,
Jason.
More information about the devicetree-discuss
mailing list