[PATCH v3 4/8] phy: rockchip-usb: add compatible values for rk3066a and rk3188

Doug Anderson dianders at chromium.org
Wed Nov 25 10:35:54 PST 2015


Hi,

On Wed, Nov 25, 2015 at 10:24 AM, Heiko Stübner <heiko at sntech.de> wrote:
> Am Mittwoch, 25. November 2015, 09:04:19 schrieb Doug Anderson:
>> Hi,
>>
>> On Sun, Nov 22, 2015 at 11:49 AM, Heiko Stuebner <heiko at sntech.de> wrote:
>> > Am Donnerstag, 19. November 2015, 16:32:23 schrieb Doug Anderson:
>> >> Heiko,
>> >>
>> >> On Thu, Nov 19, 2015 at 1:22 PM, Heiko Stuebner <heiko at sntech.de> wrote:
>> >> > We need custom handling for these two socs in the driver shortly,
>> >> > so add the necessary compatible values to binding and driver.
>> >> >
>> >> > Signed-off-by: Heiko Stuebner <heiko at sntech.de>
>> >> > ---
>> >> >
>> >> >  Documentation/devicetree/bindings/phy/rockchip-usb-phy.txt | 5 ++++-
>> >> >  drivers/phy/phy-rockchip-usb.c                             | 2 ++
>> >> >  2 files changed, 6 insertions(+), 1 deletion(-)
>> >> >
>> >> > diff --git a/Documentation/devicetree/bindings/phy/rockchip-usb-phy.txt
>> >
>> > b/Documentation/devicetree/bindings/phy/rockchip-usb-phy.txt
>> >
>> >> > index 826454a..9b37242 100644
>> >> > --- a/Documentation/devicetree/bindings/phy/rockchip-usb-phy.txt
>> >> > +++ b/Documentation/devicetree/bindings/phy/rockchip-usb-phy.txt
>> >> > @@ -1,7 +1,10 @@
>> >> >
>> >> >  ROCKCHIP USB2 PHY
>> >> >
>> >> >  Required properties:
>> >> > - - compatible: rockchip,rk3288-usb-phy
>> >> > + - compatible: matching the soc type, one of
>> >> > +     "rockchip,rk3066a-usb-phy"
>> >> > +     "rockchip,rk3188-usb-phy"
>> >> > +     "rockchip,rk3288-usb-phy"
>> >>
>> >> I can never quite keep it straight how this is supposed to work, but
>> >> since previously only "rockchip,rk3288-usb-phy" was supported and now
>> >> we have these new compatible strings, I would have expected the new
>> >> strings to specify the old ones as fallback.  That would mean your
>> >> choices would be:
>> >>
>> >> - "rockchip,rk3288-usb-phy" - A real rk3288
>> >> - "rockchip,rk3188-usb-phy", "rockchip,rk3288-usb-phy" - A rk3188 with
>> >> fallback to 3288 driver.
>> >> - "rockchip,rk3066a-usb-phy", "rockchip,rk3288-usb-phy" - A rk3066a
>> >> with fallback to 3288 driver.
>> >
>> > How this is supposed to be done also is sometimes confusing for me :-)
>> >
>> > But I don't think that specifying the "fallbacks" is part of the binding
>> > at
>> > all, when the binding really is done in a soc-specific way.  For example
>> > following the suggestion of the dt-maintainers at the time we're
>> > specifying
>> > the uarts as
>> >
>> >         compatible = "rockchip,rk3288-uart", "snps,dw-apb-uart"
>> >
>> > as a measure to use a more-special driver if there is ever the need for
>> > it.
>> > But here the "snps,dw-apb-uart" actually is a superset (a more generic
>> > implementation), while in the usb-uart-case
>>
>> Hrm, this gets into the whole issue of coming up with generic names.
>> It's not always easy, especially when marketing gets involved.  If
>> Synopsis comes up with a new APB UART that's not compatible, I guess
>> you call it a v2 and people just need to figure out which one they
>> have?  I remember it being terribly confusing with exynos since
>> Samsung called things "exynos" that were very different, and I think
>> even "exynos5" devices were pretty different form each other.  Anyway,
>> that's getting pretty far afield.
>>
>> The general way of doing things in Linux is that the first driver
>> there becomes the generic, right?  So if the first supported SoC using
>> this PHY was rk3288 then it gets the name and becomes the generic. If
>> rk3066a and 3188 are 90% the same and initially can actually use the
>> same driver, then they specify the specific "3188" and the generic
>> "3288" as a fallback.  It sounds like that was what was actually done
>> in the DTS files anyway, which is right as far as I'm concerned.
>>
>> ...but personally I'd love to see it documented.  ...someone reading
>> the binding should be able to create a DTS and it's not obvious from
>> the DTS that "rk3288" is the generic as far as this binding is
>> concerned.
>
> I'd like to disagree here :-)
>
> Generic is actually currently "rockchip-usb-phy", the platform-driver name,
> but thankfully that hasn't leaked into the dts, as even that name + filename
> should change in the future. What rk3288 (and before) uses is a Designware
> picophy with custom registers in the grf, so it should actually be
> "rockchip-usb-picophy" or so. Following socs use a different IP (Innosilicon
> if I remember correctly), with different clock/pll handling and a whole
> different set of registers, so will get a new driver.
>
> In terms of hardware compatibility, the phys aren't actually compatible, it's
> only per chance that the used SIDDQ bit has the same position on all socs :-)
> Everything else seems to move around quite happily in the registers.
>
>
> As it stands now, the rk3188 dts has a compatible of
> "rockchip,rk3188-usb-phy", "rockchip,rk3288-usb-phy", matching the usable
> [refraining from calling it generic] driver on old kernels, while now it is
> supposed to match the actually correct rk3188 variant. So that combination
> works with the same dts on both old and new kernels fullfilling the ABI-
> stability promise.
>
> With the new matching code (clock-names etc) you actually get issues if you
> try to match against "rockchip,rk3288-usb-phy" on a rk3188, so there is no
> "generic" part. And that difference will widen if we need to control other
> parts of the usb-phy as well.
>
> So essentially that second property should go completely, I just didn't wanted
> to create the impression of changing the ABI here ;-)

OK, fair enough.  I'm not terribly familiar with rk3066a and rk3188,
so didn't realize how different they were.  I guess I assumed that
they were more like 90% compatible, but perhaps not...

I also think some of these corner cases of device tree still haven't
seeped into my consciousness yet and become instinct.  Maybe in
another few years.  ;)

-Doug



More information about the linux-arm-kernel mailing list