[PATCH 13/22] drm: bridge: dw-hdmi: Replace device type with platform quirks

[Laurent Pinchart]

Hi Jose,

On Monday 05 Dec 2016 10:50:19 Jose Abreu wrote:
> On 02-12-2016 15:43, Laurent Pinchart wrote:
> > On Friday 02 Dec 2016 14:24:01 Russell King - ARM Linux wrote:
> >> On Fri, Dec 02, 2016 at 01:43:28AM +0200, Laurent Pinchart wrote:
> >>> From: Kieran Bingham <kieran.bingham+renesas at ideasonboard.com>
> >>> 
> >>> The dw-hdmi driver declares a dev_type to distinguish platform specific
> >>> changes. Replace this with a quirk field, so that the platform can
> >>> specify the required quirks for the driver, rather than the driver
> >>> becoming conditional on multiple platforms.
> >>> 
> >>> As part of this, we rename the dw-hdmi 'spare' which is defined as the
> >>> SVSRET bit in later documentation.
> >> 
> >> I'd really prefer that we did not go down the broken route of adding
> >> a set of "quirk" flags - look at what a mess SDHCI has become through
> >> allowing that kind of practice.
> >> 
> >> I'd much rather we find a saner structure to this - and we know that
> >> the hardware has ID registers in it which can be used (so far) to
> >> identify the buggy hardware.
> > 
> > I'd much prefer something that would allow runtime identification of the
> > device and the corresponding actions to be taken. However, the amount of
> > documentation we have on the DWC HDMI TX IP core (and the associated PHY)
> > is pretty limited, given that Synopsys doesn't make the documentation
> > available publicly. Changes made to the IP core by integrators could
> > complicate this further. I'm trying to gather as much information as
> > possible to make clean the code up, for instance by trying to identify
> > the PHYs used on the various platforms we support. Progress is slow on
> > that front, there isn't enough leaked information available online :-) I
> > haven't given up though, but I'll need more time.
> > 
> > I don't like quirks much either. They are however already used today, even
> > if we trigger them through dev_type instead of quirk flags. This patch
> > came from a previous version found in a BSP that simply sprinkled several
> > if (hdmi-> dev_type == RCAR_HDMI) through the code. For instance,
> > 
> > -	if (hdmi->dev_type == RK3288_HDMI)
> > +	if (hdmi->dev_type == RK3288_HDMI || hdmi->dev_type == RCAR_HDMI)
> > 		dw_hdmi_phy_enable_spare(hdmi, 1);
> > 
> > which I think is worse than flags as it would quickly degenerate to
> > spaghetti code.
> > 
> > For this specific case, we've managed to identify that on Renesas
> > platforms the bit set by this function is called SVSRET. Its usage isn't
> > clear yet, but I suspect it to control one of the PHY input control
> > signals, like the other bits in the same register. I'm trying to get more
> > information to clean the implementation further, hopefully with a way to
> > determine whether the signal is used based on PHY identification.
> 
> SVSRET is a low power mode consumption and is a PHY input signal
> as you suggested.

Thank you for the confirmation. Would you happen to know what SVSRET stands 
for ?

> Most of the configurable input signals of the PHY are available by the
> controller regbank. I don't think it is possible to detect this at runtime,
> I think you have at least to hardcode which version of the PHY you are
> using.
>
> I would suggest that maybe all the PHY logic should be extracted and then
> use callbacks to glue controller and phy. Then, depending on the PHY you
> could use empty stubs if, for example, a given PHY did not support SVSRET.
> Still, I don't know if this is the best option. What I do know is that there
> are a large number of PHY's with different flavors that can use the same
> controller. The controller has different versions also, and each version can
> have quirks but I think it would be easier to manage this driver if we had a
> clear distinction between PHY and controller.

Agreed, I'd like to go in that direction. What makes it quite difficult is the 
lack of documentation about the PHYs :-) I've found six different PHY types 
that can be identified by the CONFIG2_ID register:

Bits    | Field         	| Description
------------------------------------------------------------------------------
7-0     | phytype       	| PHY interface
        |               	| 0x00: Legacy PHY (HDMI TX PHY)
        |               	| 0xb2: MHL PHY + HEAC PHY
        |               	| 0xc2: MHL PHY
        |               	| 0xe2: HDMI 3D TX PHY + HEAC PHY
        |               	| 0xf2: HDMI 3D TX PHY
        |               	| 0xf3: HDMI2 TX PHY

I'm sure there's more than that. In particular I wonder how external vendor 
PHYs are identified.

I'm also wondering whether there's a need to keep support for the legacy PHY 
signals (ENTMDS and PDZ in the PHY_CONF0 register). As far as I understand 
they're not used by the Gen2 PHYs (including the external vendor PHYs), but I 
can't confirm that without more documentation (although I could test that on 
the platforms I have access to).

> > This is all work in progress, and if anyone has access to any
> > documentation and can provide additional information I'll be grateful.
> > 
> >>> Signed-off-by: Kieran Bingham <kieran.bingham+renesas at ideasonboard.com>
> >>> Signed-off-by: Laurent Pinchart
> >>> <laurent.pinchart+renesas at ideasonboard.com>
> >>> ---
> >>> 
> >>>  drivers/gpu/drm/bridge/dw-hdmi.c            | 14 ++++++--------
> >>>  drivers/gpu/drm/bridge/dw-hdmi.h            |  4 ++--
> >>>  drivers/gpu/drm/imx/dw_hdmi-imx.c           |  3 +--
> >>>  drivers/gpu/drm/rockchip/dw_hdmi-rockchip.c |  2 +-
> >>>  include/drm/bridge/dw_hdmi.h                | 12 +++++-------
> >>>  5 files changed, 15 insertions(+), 20 deletions(-)

-- 
Regards,

Laurent Pinchart