[PATCH v3 41/50] drm/bridge: ti-tfp410: Report input bus config through bridge timings

[Laurent Pinchart]

Hi Tomi,

On Fri, Mar 15, 2019 at 03:30:11PM +0200, Tomi Valkeinen wrote:
> On 15/03/2019 14:28, Peter Ujfalusi wrote:
> > On 15/03/2019 14.07, Tomi Valkeinen wrote:
> >>> If the pclk-sample is not defined in DT, it will default to 0 which
> >>> selects SAMPLE_NEGEDGE (== DRIVE_POSEDGE), right?
> >>>
> >>> But all the boards where I can find schematics with tfp410 have their
> >>> EDGE/HTPLG pin pulled up and according to the documentation when EDGE=1
> >>> then tfp410 will sample on the rising edge.
> >>>
> >>> imho the pclk_sample should be initialized to 1 to avoid regression for
> >>> most of the boards using tfp410.
> >>
> >> Define "regression" =). If the omapdrm driver was always using
> >> DRIVE_POSEDGE, this driver should also be using DRIVE_POSEDGE, no? If it
> >> does the same as the old driver, it can't be a regression. This is, of
> >> course, only considering omapdrm based boards.
> >>
> >> That said, it sounds odd that this would be wrong in the old driver, but
> >> then again, it might well be, as code related to these sync signals has
> >> changed sooo many times, and the related DSS registers are somewhat
> >> confusing.
> > 
> > On the boards data skew is enabled as well with maximum delay selected
> > with DK1=DK2=DK3=1, so sampling of data is delayed by 3 * 350 ps, so
> > about 1 ns, not much, but might be enough for the signal to transition
> > on the bus so even if the HW drivers on POSEDGE and tfp410 samples on
> > POSEDGE (+1ns delay from the edge) we don't see corruption?
> 
> Ok. I don't think anyone has looked at it that closely. So, the syncs
> could well be wrong. Still, not a regression if it's the same way as it was.
> 
> If the fixed default works fine (or better) than the wrong ones, I think
> that can be the default. But I'd be careful about changing what the
> default is, if we've had it the old way for a long time.

I don't object changing the default (given proper testing), but that
should be done on top of this series then.

-- 
Regards,

Laurent Pinchart