Re: [RFC PATCH 03/10] drm/mipi-dsi: add API for manual control over the DSI link power state

[Dmitry Baryshkov]

On Mon, 23 Oct 2023 at 09:52, Alexander Stein
<alexander.stein@xxxxxxxxxxxxxxx> wrote:
>
> Hi Dmitry,
>
> Am Sonntag, 22. Oktober 2023, 12:49:41 CEST schrieb Dmitry Baryshkov:
> > On Thu, 19 Oct 2023 at 14:42, Alexander Stein
> >
> > <alexander.stein@xxxxxxxxxxxxxxx> wrote:
> > > Hi,
> > >
> > > Am Donnerstag, 19. Oktober 2023, 13:19:51 CEST schrieb Dmitry Baryshkov:
> > > > On Thu, 19 Oct 2023 at 12:26, Maxime Ripard <mripard@xxxxxxxxxx> wrote:
> > > > > On Mon, Oct 16, 2023 at 07:53:48PM +0300, Dmitry Baryshkov wrote:
> > > > > > The MIPI DSI links do not fully fall into the DRM callbacks model.
> > > > >
> > > > > Explaining why would help
> > > >
> > > > A kind of explanation comes afterwards, but probably I should change
> > > > the order of the phrases and expand it:
> > > >
> > > > The atomic_pre_enable / atomic_enable and correspondingly
> > > > atomic_disable / atomic_post_disable expect that the bridge links
> > > > follow a simple paradigm: either it is off, or it is on and streaming
> > > > video. Thus, it is fine to just enable the link at the enable time,
> > > > doing some preparations during the pre_enable.
> > > >
> > > > But then it causes several issues with DSI. First, some of the DSI
> > > > bridges and most of the DSI panels would like to send commands over
> > > > the DSI link to setup the device. Next, some of the DSI hosts have
> > > > limitations on sending the commands. The proverbial sunxi DSI host can
> > > > not send DSI commands after the video stream has started. Thus most of
> > > > the panels have opted to send all DSI commands from pre_enable (or
> > > > prepare) callback (before the video stream has started).
> > > >
> > > > However this leaves no good place for the DSI host to power up the DSI
> > > > link. By default the host's pre_enable callback is called after the
> > > > DSI bridge's pre_enable. For quite some time we were powering up the
> > > > DSI link from mode_set. This doesn't look fully correct. And also we
> > > > got into the issue with ps8640 bridge, which requires for the DSI link
> > > > to be quiet / unpowered at the bridge's reset time.
> > >
> > > There are also bridges (e.g. tc358767) which require DSI-LP11 upon bridge
> > > reset. And additionally this DSI-(e)DP bridge requires LP11 while
> > > accessing
> > > DP-AUX channel, e.g. reading EDID. So bridges need at least some control
> > > over DSI line state.
> >
> > For sending commands in LP11 it is typical to toggle the
> > MIPI_DSI_MODE_LPM flag, for example see panel-=jdi-lt070me05000.c or
> > some other drives. It might be a good idea to make that more explicit.
> > All suggestions here would be appreciated.
>
> The biggest difference between that display and the tc358767 bridge is that
> the display uses DSI commands, while the bridge is using i2c transfer to issue
> DP-AUX commands. There is no host_transfer [1] which would enable LP-11.
> It seems this DSI-DP bridge requires LP-11/HS on DSI lanes all times. This
> contradicts current Linux behaviour.

I see. I took a quick glance at the driver. Does the device mark AUX
as busy when there is a HS transfer?
Because otherwise it might be pretty hard to synchronise DP-AUX
transfers with the DSI link state. We might need to add an API for
this, if the DSI hosts actually can signal the blanking / DSI LP.

Side note: the driver needs some care. It doesn't support the aux-bus
bindings for eDP panels, it doesn't support other bridges on top of DP
connectors (but there can be e..g. dp-connector device).

>
> Best regards,
> Alexander
>
> [1] https://www.kernel.org/doc/html/latest/gpu/drm-kms-helpers.html#mipi-dsi-bridge-operation


-- 
With best wishes
Dmitry