[PATCH 4/4 v2] ARM: S5PV210: Add clock definition for MIPI-CSIS DPHY

[Marek Szyprowski]

Hello,

On Friday, December 10, 2010 1:01 PM Sylwester Nawrocki wrote:

> On 2010-12-10 10:12, Marek Szyprowski wrote:
> > On Friday, December 10, 2010 6:14 AM Kukjin Kim wrote:
> >> Sylwester Nawrocki wrote:
> >>>
> >>> MIPI DPHY control register requires special handling since it is shared
> >>> between CSI (camera serial interface) and DSI (display serial interface).
> >>> By creating this clock a serialized interface is provided for mipi-csis
> >>> and mipi-dsim drivers, so DPHYs may be safely controlled by both drivers.
> >>> Similarly dsim_dphy clock could be added for mipi-dsim.
> >>>
> >>> ---
> >>>
> >>> I am not quite sure about_"dphy_clock", perhaps power domain
> >>> handling code would be better place for it.
> >>>
> >> Yeah, it is MIPI DPHY enable/disable control register not clock control.
> >> So its proper position is not here...
> >>
> >> Hmm...how about driver's probe/open or some kind of setup in machine
> >> directory?
> >
> > I'm not sure that this is the best way of handling this phy controller. Please
> > notice that mipi phy controller has different register location on S5PC110 and
> > S5PC210. Please also notice that S5PC210 has 2 mipi csci controllers and phy
> > interfaces, while s5pc110 has only one. Hiding all the logic and register
> > specific writes behind this 'csi_dphy' clock easily resolves all these issues
> > on all different samsung platforms and makes it easy to use it from the driver.
> > Similar patch has been proposed some time ago for usb_phy interface and imho
> > this is the right way to go.
> >
> More importantly S5P_MIPI_DPHY_CONTROL* register is shared between two
> devices - MIPI-CSIS and MIPI-DSIM. It means some kind of serialization
> mechanism need to be employed while accessing this register, a spinlock
> seems to fit best here. It also means that if we decided to create the
> platform callback such a code would have to be placed in a common
> compilation unit which would be built when either mipi-csis or mipi-dsim
> is selected. And this callback would have to account the differences
> between various SoC flavours or it would have to be repeated for each
> mach-s5pvXXX.
> So if the platform callback is the preferred way, where do we put an
> implementation of it?

That's yet another advantage of clock-based approach. All clock related operations
are automatically serialized by global clock API spinlock.

Best regards
--
Marek Szyprowski
Samsung Poland R&D Center