Re: [PATCH v12 09/11] media: staging: dt-bindings: add Rockchip MIPI RX D-PHY yaml bindings

[Ezequiel Garcia <ezequiel@collabora.com>]

On Tue, 2020-01-07 at 23:25 +0100, Heiko Stuebner wrote:
> Am Dienstag, 7. Januar 2020, 23:03:54 CET schrieb Ezequiel Garcia:
> > On Tue, 2020-01-07 at 22:30 +0100, Heiko Stübner wrote:
> > > Hi Ezequiel,
> > > 
> > > Am Dienstag, 7. Januar 2020, 14:20:10 CET schrieb Ezequiel Garcia:
> > > > Hi Heiko, Laurent,
> > > > 
> > > > On Tue, 2020-01-07 at 10:28 +0100, Heiko Stübner wrote:
> > > > > Am Dienstag, 7. Januar 2020, 03:37:21 CET schrieb Laurent Pinchart:
> > > > > > On Mon, Jan 06, 2020 at 11:06:12PM -0300, Ezequiel Garcia wrote:
> > > > > > > On Tue, 2020-01-07 at 02:10 +0200, Laurent Pinchart wrote:
> > > > > > > > Hi Helen,
> > > > > > > > 
> > > > > > > > Thank you for the patch.
> > > > > > > > 
> > > > > > > > On Fri, Dec 27, 2019 at 05:01:14PM -0300, Helen Koike wrote:
> > > > > > > > > Add yaml DT bindings for Rockchip MIPI D-PHY RX
> > > > > > > > > 
> > > > > > > > > This was tested and verified with:
> > > > > > > > > mv drivers/staging/media/phy-rockchip-dphy/Documentation/devicetree/bindings/phy/rockchip-mipi-
> > > > > > > > > dphy.yaml  Documentation/devicetree/bindings/phy/
> > > > > > > > > make dt_binding_check DT_SCHEMA_FILES=Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > make dtbs_check DT_SCHEMA_FILES=Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > 
> > > > > > > > > Signed-off-by: Helen Koike <helen.koike@collabora.com>
> > > > > > > > > 
> > > > > > > > > ---
> > > > > > > > > 
> > > > > > > > > Changes in v12:
> > > > > > > > > - The commit replaces the following commit in previous series named
> > > > > > > > > media: staging: dt-bindings: Document the Rockchip MIPI RX D-PHY bindings
> > > > > > > > > This new patch adds yaml binding and was verified with
> > > > > > > > > make dtbs_check and make dt_binding_check
> > > > > > > > > 
> > > > > > > > > Changes in v11: None
> > > > > > > > > Changes in v10:
> > > > > > > > > - unsquash
> > > > > > > > > 
> > > > > > > > > Changes in v9:
> > > > > > > > > - fix title division style
> > > > > > > > > - squash
> > > > > > > > > - move to staging
> > > > > > > > > 
> > > > > > > > > Changes in v8: None
> > > > > > > > > Changes in v7:
> > > > > > > > > - updated doc with new design and tested example
> > > > > > > > > 
> > > > > > > > >  .../bindings/phy/rockchip-mipi-dphy.yaml      | 75 +++++++++++++++++++
> > > > > > > > >  1 file changed, 75 insertions(+)
> > > > > > > > >  create mode 100644 drivers/staging/media/phy-rockchip-dphy/Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > 
> > > > > > > > > diff --git a/drivers/staging/media/phy-rockchip-dphy/Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > b/drivers/staging/media/phy-
> > > > > > > > > rockchip-dphy/Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > new file mode 100644
> > > > > > > > > index 000000000000..af97f1b3e005
> > > > > > > > > --- /dev/null
> > > > > > > > > +++ b/drivers/staging/media/phy-rockchip-dphy/Documentation/devicetree/bindings/phy/rockchip-mipi-dphy.yaml
> > > > > > > > > @@ -0,0 +1,75 @@
> > > > > > > > > +# SPDX-License-Identifier: (GPL-2.0+ OR MIT)
> > > > > > > > > +%YAML 1.2
> > > > > > > > > +---
> > > > > > > > > +$id: http://devicetree.org/schemas/phy/rockchip-mipi-dphy.yaml#
> > > > > > > > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > > > > > > > +
> > > > > > > > > +title: Rockchip SoC MIPI RX0 D-PHY Device Tree Bindings
> > > > > > > > 
> > > > > > > > Should this be s/RX0/RX/ ? Or do you expect different bindings for RX1 ?
> > > > > > > 
> > > > > > > The driver currently only supports RX0, but I think you are right,
> > > > > > > it should say RX here. This binding could be extended for RX1.
> > > > > > > 
> > > > > > > > Looking at the PHY driver, it seems to handle all PHYs with a single
> > > > > > > > struct device. Should we thus use #phy-cells = <1> to select the PHY ?
> > > > > > > 
> > > > > > > I am not following this. The driver handles just one PHY. Each PHY
> > > > > > > should have its own node.
> > > > > > 
> > > > > > Looking at the registers, it seems that the different PHYs are
> > > > > > intertwined and we would could have trouble handling the different PHYs
> > > > > > with different DT nodes and thus struct device instances.
> > > > > 
> > > > > I have to confess to not following _ALL_ of the threads, so may say
> > > > > something stupid, but I don't think the PHYs are intertwined so much.
> > > > > 
> > > > > Where RX0 is controlled from the "General Register Files" alone
> > > > > [register dumping ground for soc designers], the TX1RX1-phy
> > > > > actually gets controlled from inside the dsi1 register area it seems.
> > > > > 
> > > > > So in my previous (still unsucessful) tests, I was rolling with something like
> > > > > https://github.com/mmind/linux-rockchip/commit/e0d4b03976d2aab85a8c1630be937ea003b5df88
> > > > > 
> > > > > With the actual "logic" picked from the vendor kernel, that just double-
> > > > > maps the dsi1-registers in both dsi and dphy driver, which was strange.
> > > > > 
> > > > > 
> > > > 
> > > > Describing each PHY in its own device node (as we currently do)
> > > > results in:
> > > > 
> > > >         mipi_dphy_tx1rx1: mipi-dphy-tx1rx1@ff968000 {
> > > >                 compatible = "rockchip,rk3399-mipi-dphy";
> > > >                 reg = <0x0 0xff968000 0x0 0x8000>;
> > > >                 rockchip,grf = <&grf>;
> > > >         };
> > > 
> > > 0xff968000 actually really is the dsi1 controller, so we'll already
> > > have a node for that area. That is the reason I went that way to make
> > > the rockchip-dsi optionally also behave as phy-provider.
> > > 
> > > So when it's used in combination with drm and a panel or so it will
> > > behave as dsi controller, but when requested via the phy-framework
> > > it will expose the dphy functionality.
> > > 
> > 
> > Hm, and will this driver also support RX1?
> 
> what is RX1 in your book? :-)
> 

I meant tx1rx1 configured as a receiver,
sorry for the confusion!

> According to the TRM the rk3399 has 3 DPHYs,
> tx0 - connected exclusively to dsi0
>       (this is handled internally by the dw-mipi-dsi driver with controls
>        in the dsi0 register space)
> rx0 - connected exclusively to isp0
>       (this is handled by the individual dphy driver from Helen's series)
> tx1rx1 - shared between dsi1 and isp1
>       (again inside the dsi1 register space)
> 

Exactly.

> 
> > > >         grf: syscon@ff770000 {
> > > >                 mipi_dphy_rx0: mipi-dphy-rx0 {
> > > >                         compatible = "rockchip,rk3399-mipi-dphy";
> > > >                 };
> > > >         };
> > > > 
> > > > Which is mildly ugly, as it uses two mechanism to describe
> > > > the GRF resource. In addition, the driver will then _infer_
> > > > which device node is RX0 and which is TX1RX1, from this.
> > > > 
> > > > Perhaps Laurent's proposal, describing each PHY explicitly,
> > > > would be cleaner?
> > > 
> > > so I really think we shouldn't merge these two things together,
> > > especially to not break the dsi1 controller part.
> > > 
> > 
> > I don't think it would necesarily break the dsi1 controller part.
> > 
> > You can declare both device nodes as sharing the address region,
> > and then the driver can request the I/O resource only when it needs to,
> > i.e. in the PHY .init hook.
> 
> dsi1 is of course a dw-mipi-dsi one, which in turn shares a common bridge
> driver over multiple variants (non-rockchip), which expects its registers
> mapped during probe.
> 
> I think it would not really work well if you need to make the whole world
> follow that idea ;-) .
> 
> 
> Hence my approach with exposing the phy interface from the dsi driver.
> If you look at the dts part, it also just looks like it should be ... as
> a regular phy:
> 	https://github.com/mmind/linux-rockchip/blob/wip/tc358749/arch/arm64/boot/dts/rockchip/rk3399.dtsi#L1764
> 
> And on the driver side there is even some short circuit protection.
> When used as phy, it won't allow to be used as a component and
> vice versa.
> 

OK, it's more clear now.

If this bindings will only ever support the RX0 PHY,
will not support anything else, then we don't need
to worry.

It can stay as a GRF syscon child, and will have
phy-cells = 0.

Thanks,
Ezequiel


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel