On Mon, Mar 26, 2018 at 04:40:21PM +0200, Enric Balletbo Serra wrote:

Hi Pavel,

2018-03-26 12:42 GMT+02:00 Pavel Machek [off-list ref]:

quoted

On Mon 2018-03-26 11:52:09, Enric Balletbo i Serra wrote:

quoted

Before this patch the enable signal was set before the PWM signal and
vice-versa on power off. This sequence is wrong, at least, it is on
the different panels datasheets that I checked, so I inverted the sequence
to follow the specs.

For reference the following panels have the mentioned sequence:
  - N133HSE-EA1 (Innolux)
  - N116BGE (Innolux)
  - N156BGE-L21 (Innolux)
  - B101EAN0 (Auo)
  - B101AW03 (Auo)
  - LTN101NT05 (Samsung)
  - CLAA101WA01A (Chunghwa)

Ok, but this changes behaviour for other panels, too. Are you sure you
are not breaking one of those?

I can't say that I am 100% sure because I didn't find all the
datasheets of all the panels supported in the kernel. But all the
datasheets I checked specifies this sequence as valid. In general I
think that doesn't really matter, but I know that at least the
B116XTN02 panel requires enable first the PWM, wait 10ms and then
enable BL_EN to avoid garbage. So the other way around is not valid
for this panel. That's the reason for this patchset.

This is certainly a patch that could cause regressions... but it would
be a very odd panel that *likes* to be exposed to all the weird edges
that might occur whilst the PWM stablizes and a panel that *needs* to
see weird edges to work seems even less likely (since it could act
different with each SoC).

So whilst the patch is not absolutely cast iron guaranteed free of
risk, I think it is well enough argued for.

Naturally I may change my position *very* quickly on receipt of the
first bug report ;-)


Daniel.