On 2023-05-30 14:11:06, Dmitry Baryshkov wrote:
> On Tue, 30 May 2023 at 11:27, Marijn Suijten
> <marijn.suijten@xxxxxxxxxxxxxx> wrote:
> >
> > On 2023-05-30 01:39:10, Dmitry Baryshkov wrote:
> > > On 30/05/2023 01:37, Marijn Suijten wrote:
> > > > On 2023-05-30 01:18:40, Dmitry Baryshkov wrote:
> > > > <snip>
> > > >>>>>>> + ret = mipi_dsi_dcs_set_display_on(dsi);
> > > >>>>>>> + if (ret < 0) {
> > > >>>>>>> + dev_err(dev, "Failed to turn display on: %d\n", ret);
> > > >>>>>>> + return ret;
> > > >>>>>>> + }
> > > >>>>>>
> > > >>>>>> My usual question: should the mipi_dsi_dcs_exit_sleep_mode() / mipi_dsi_dcs_set_display_on() be moved from prepare() to enable() part?
> > > >>>>>
> > > >>>>>
> > > >>>>> No, prepare is called before the video stream is started and when display is still in LPM mode and the mode hasn't been set.
> > > >>>>>
> > > >>>>
> > > >>>> Yes, that's my point. Shouldn't we enable the panel _after_ starting the stream?
> > > >>>
> > > >>> I have never investigated what it takes to split these functions, but
> > > >>> some of these panels do show some corruption at startup which may be
> > > >>> circumvented by powering the panel on after starting the video stream?
> > > >>>
> > > >>> I'm just not sure where to make the split: downstream does describe a
> > > >>> qcom,mdss-dsi-on-command and qcom,mdss-dsi-post-panel-on-command, where
> > > >>> the latter only contains set_display_on() (not exit_sleep_mode()).
> > > >>> It is documented like:
> > > >>>
> > > >>> same as "qcom,mdss-dsi-on-command" except commands are sent after
> > > >>> displaying an image."
> > > >>>
> > > >>> So this seems like the right way to split them up, I'll test this out on
> > > >>> all submitted panel drivers.
> > > >>
> > > >> Interesting enough, Neil suggested that sending all the commands during
> > > >> pre_enable() is the correct sequence (especially for VIDEO mode panels),
> > > >> since not all DSI hosts can send commands after switching to the VIDEO mode.
> > > >
> > > > Note that all these panels and Driver-ICs are command-mode, and/or
> > > > programmed to run in command-mode, so there shouldn't be any notion of a
> > > > VIDEO stream (any command-mode frame is just an "arbitrary command" as
> > > > far as I understood).
> > >
> > > Yes, from the data stream point of view. I was talking about the DSI
> > > host being able to send arbitrary commands or not after enabling the
> > > video/cmd stream.
> >
> > Is this a known limitation of SM8250 then? Or is the brightness_set
> > issue more likely a "problem" with the panel that the downstream kernel
> > is "somehow" working around or aware of, and I just haven't found
> > how/where it deals with that?
> > (Alternatively we could be "doing it wrong" for other panels but it
> > turns out to be working anyway)
>
> Please excuse me for not being explicit enough. Qualcomm hardware
> doesn't have this problem. Thus I was completely unaware of it before
> talking to Neil.
> So, our hardware works in most of the cases.
Also excuse me for mocking the hardware here; it seems quite illogical
for it to not work on this specific device which is more likely a
failure in porting the panel DT to the driver than related to this
specific SoC. There's probably one of the hundred-or-so DT params
responsible for triggering a setting, delay, or other magic sequence
that gets the brightness toggle working.
- Marijn
