[linux-sunxi] Re: [RFC PATCH 07/11] drm: sun4i: add support for the TV encoder in H3 SoC

[icenowy at aosc.io]

在 2017-06-01 02:43，Maxime Ripard 写道：
> On Wed, May 24, 2017 at 04:25:46PM +0800, Icenowy Zheng wrote:
>> 
>> 
>> 于 2017年5月24日 GMT+08:00 下午3:30:19, Maxime Ripard 
>> <maxime.ripard at free-electrons.com> 写到:
>> >On Tue, May 23, 2017 at 09:00:59PM +0800, icenowy at aosc.io wrote:
>> >> 在 2017-05-23 20:53，Maxime Ripard 写道：
>> >> > On Mon, May 22, 2017 at 07:55:56PM +0200, Jernej Škrabec wrote:
>> >> > > Hi,
>> >> > >
>> >> > > Dne sobota, 20. maj 2017 ob 03:37:53 CEST je Chen-Yu Tsai
>> >napisal(a):
>> >> > > > On Sat, May 20, 2017 at 2:23 AM, Jernej Škrabec
>> ><jernej.skrabec at siol.net>
>> >> > > wrote:
>> >> > > > > Hi,
>> >> > > > >
>> >> > > > > Dne petek, 19. maj 2017 ob 20:08:18 CEST je Icenowy Zheng
>> >napisal(a):
>> >> > > > >> 于 2017年5月20日 GMT+08:00 上午2:03:30, Maxime Ripard
>> ><maxime.ripard at free-
>> >> > > > >
>> >> > > > > electrons.com> 写到:
>> >> > > > >> >On Thu, May 18, 2017 at 12:43:50AM +0800, Icenowy Zheng
>> >wrote:
>> >> > > > >> >> Allwinner H3 features a TV encoder similar to the one in
>> >earlier
>> >> > > > >> >
>> >> > > > >> >SoCs,
>> >> > > > >> >
>> >> > > > >> >> but with some different points about clocks:
>> >> > > > >> >> - It has a mod clock and a bus clock.
>> >> > > > >> >> - The mod clock must be at a fixed rate to generate
>> >signal.
>> >> > > > >> >
>> >> > > > >> >Why?
>> >> > > > >>
>> >> > > > >> It's experiment result by Jernej.
>> >> > > > >>
>> >> > > > >> The clock rates in BSP kernel is also specially designed
>> >> > > > >> (PLL_DE at 432MHz) in order to be able to feed the TVE.
>> >> > > > >
>> >> > > > > My experiments and search through BSP code showed that TVE
>> >seems to have
>> >> > > > > additional fixed predivider 8. So if you want to generate 27
>> >MHz clock,
>> >> > > > > unit has to be feed with 216 MHz.
>> >> > > > >
>> >> > > > > TVE has only one PLL source PLL_DE. And since 216 MHz is a
>> >bit low for
>> >> > > > > DE2,
>> >> > > > > BSP defaults to 432 MHz for PLL_DE and use divider 2 to
>> >generate 216 MHz.
>> >> > > > > This clock is then divided by 8 internaly to get final 27
>> >MHz.
>> >> > > > >
>> >> > > > > Please note that I don't have any hard evidence to support
>> >that, only
>> >> > > > > experimental data. However, only that explanation make sense
>> >to me.
>> >> > > > >
>> >> > > > > BTW, BSP H3/H5 TV driver supports only PAL and NTSC which
>> >both use 27 MHz
>> >> > > > > base clock. Further experiments are needed to check if there
>> >is any
>> >> > > > > possibility to have other resolutions by manipulating clocks
>> >and give
>> >> > > > > other proper settings. I plan to do that, but not in very
>> >near future.
>> >> > > >
>> >> > > > You only have composite video output, and those are the only 2
>> >standard
>> >> > > > resolutions that make any sense.
>> >> > >
>> >> > > Right, other resolutions are for VGA.
>> >> > >
>> >> > > Anyway, I did some more digging in A10 and R40 datasheets. I
>> >think
>> >> > > that H3 TVE
>> >> > > unit is something in between. R40 TVE has a setting to select "up
>> >> > > sample".
>> >> >
>> >> > That might be just another translation of oversampling :)
>> >> >
>> >> > I didn't know it could be applied to composite signals though, but
>> >I
>> >> > guess this is just another analog signal after all.
>> >> >
>> >> > > Possible settings are 27 MHz, 54 MHz, 108 MHz and 216 MHz. BSP
>> >> > > driver on R40
>> >> > > has this setting enabled only for PAL and NTSC and it is always
>> >216
>> >> > > MHz. I
>> >> > > think that H3 may have this hardwired to 216 MHz and this would
>> >be
>> >> > > the reason
>> >> > > why 216 MHz is needed.
>> >> > >
>> >> > > Has anyone else any better explanation?
>> >> >
>> >> > That's already a pretty good one.
>> >> >
>> >> > Either way, wether this is upsampling, oversampling or just a
>> >> > pre-divider, this can and should be dealt with in the mode_set
>> >> > callback, and not in the probe.
>> >>
>> >> I got a better idea -- let TVE driver have the CLK_TVE as an
>> >> input and create a subclock output with divider 16, and feed this
>> >> subclock to TCON lcd-ch1.
>> >>
>> >> This is a model of the real hardware -- the clock divider is in
>> >> TVE, not TCON.
>> >
>> >That's definitely not a good representation of the hardware. There's
>> >one clock, it goes to the TCON, period.
>> 
>> No, I still think it goes to the TVE as:
>> 
>> 1. it's named TVE in datasheet.
> 
> Feel free to come up with a better, more sensible explanation?
> 
>> 2. Generating signal with such a low resolution but such
>> a high dotclock is not a good situation.
> 
> What? What do you mean? The pixel clock is 27MHz, I'm pretty sure we
> agree on that.

Yes, so I don't think the TCON code should set the clock to any value
rather than 27MHz.

So we should create a clock with divider in TVE, and feed it to TCON.

> 
> Maxime
> 
> --
> Maxime Ripard, Free Electrons
> Embedded Linux and Kernel engineering
> http://free-electrons.com