Re: [RFC PATCH] drm/dsi-mipi: Fix byte order of DCS set/get brightness

[Sam Ravnborg <sam@ravnborg.org>]

Hi Richard,
On Fri, Jan 13, 2023 at 04:24:23PM -0500, Richard Acayan wrote:
> On Fri, Jan 13, 2023 at 05:34:55PM +0100, Sam Ravnborg wrote:
> > Hi Richard/Daniel.
> Not sure if you saw the original commit that I linked.
> 
> This patch was written in 2017 for the Pixel 3 and Pixel 3 XL
> smartphones. I didn't contact the original author before sending this,
> since the kernel that this patch originated from is EOL.
> 
> This greeting is still fine because it would be great if the original
> author took responsibility for getting this patch applied.
> 
> >
> > On Thu, Jan 12, 2023 at 11:18:48PM -0500, Richard Acayan wrote:
> >> From: Daniel Mentz <danielmentz@google.com>
> >> 
> >> The MIPI DCS specification demands that brightness values are sent in
> >> big endian byte order. It also states that one parameter (i.e. one byte)
> >> shall be sent/received for 8 bit wide values, and two parameters shall
> >> be used for values that are between 9 and 16 bits wide.
> > It is only a few week ago someone write they needed a byteswap to use
> > the kernel provided mipi_dsi_dcs_set_display_brightness(). I did not
> > realize then that this is because it is buggy.
> I didn't even consider sending this patch when I initially saw it. I
> would have performed the byte swap in a new S6E3FA7 panel driver just
> like in SOFEF00.
> 
> >
> > Some comments on the patch:
> > - It would be nice to split it up so first patch fixes that the
> >   parameters are in big-endian. This would go for both set and get.
> The brightness is only big endian if it's 16-bit.
> 
> Most MIPI panel drivers in the kernel tree only send an 8-bit
> brightness, or none at all. For these panels, the brightness can be sent
> as a 16-bit integer in little endian, as it has the same effect as an
> 8-bit integer, with a zero byte appended. The original MIPI brightness
> functions work properly with no changes for 8-bit brightness panels
> only.
> 
> If we want separate 8-bit (original) and 16-bit functions, the 8-bit
> functions shouldn't pass the brightness in big endian. That doesn't mean
> we shouldn't change them at all.
Agree, the brightness is passed exactly as returned by
backlight_get_brightness(), and the helper take care of the correct byte
ordering.

> 
> >
> > - Second patch could then introduce the possibility to a driver to
> >   use a variant that matches the display.
> >   I would suggest to go for two functions:
> >   mipi_dsi_dcs_set_display_brightness()
> >   mipi_dsi_dcs_set_display_brightness_two()
> I'm thinking of a "_large" or "_wide" suffix instead. This would make a
> 32-bit suffix tricky though, if it is ever needed.
_large is way better than _two, and would be my choice.

> 
> >   So it is obvious when you go for the two byte version without adding a
> >   bool parameter.
> >   If the implementation is done using a helper or two independent
> >   functions are up to you.
> These get_display_brightness and set_display_brightness functions are
> already helpers. I'll have them duplicated.
> 
> >
> > The patch split would be nice, but at least we should have two functions
> > and not a bool parameter.
> Yes, that was a bad idea on my part. With separate functions for 16-bit
> brightness, there is no API change, less panel driver touching, and
> panel driver changes can happen in separate patches. Thank you for the
> suggestion.
Great, looks forward for the patches.

	Sam