Re: [PATCH v1 2/3] drm/msm/dp: decoupling dp->id out of dp controller_id at scxxxx_dp_cfg table

[Stephen Boyd]

Quoting Kuogee Hsieh (2022-06-24 14:17:50)
>
> On 6/24/2022 1:00 PM, Stephen Boyd wrote:
> > Quoting Kuogee Hsieh (2022-06-24 10:15:11)
> >> Current the index (dp->id) of DP descriptor table (scxxxx_dp_cfg[]) are tightly
> >> coupled with DP controller_id. This means DP use controller id 0 must be placed
> >> at first entry of DP descriptor table (scxxxx_dp_cfg[]). Otherwise the internal
> >> INTF will mismatch controller_id. This will cause controller kickoff wrong
> >> interface timing engine and cause dpu_encoder_phys_vid_wait_for_commit_done
> >> vblank timeout error.
> >>
> >> This patch add controller_id field into struct msm_dp_desc to break the tightly
> >> coupled relationship between index (dp->id) of DP descriptor table with DP
> >> controller_id.
> > Please no. This reverts the intention of commit bb3de286d992
> > ("drm/msm/dp: Support up to 3 DP controllers")
> >
> >      A new enum is introduced to document the connection between the
> >      instances referenced in the dpu_intf_cfg array and the controllers in
> >      the DP driver and sc7180 is updated.
> >
> > It sounds like the intent of that commit failed to make a strong enough
> > connection. Now it needs to match the INTF number as well? I can't
> > really figure out what is actually wrong, because this patch undoes that
> > intentional tight coupling. Is the next patch the important part that
> > flips the order of the two interfaces?
>
> The commit bb3de286d992have two problems,
>
> 1)  The below sc7280_dp_cfg will not work, if eDP use
> MSM_DP_CONTROLLER_2 instead of  MSM_DP_CONTROLLER_1

Why would we use three indices for an soc that only has two indices
possible? This is not a real problem?

>
> since it have num_descs =2 but eDP is at index 2 (CONTROLLER_2) which
> never be reached.
>
> static const struct msm_dp_config sc7280_dp_cfg = {
>          .descs = (const struct msm_dp_desc[]) {
>                  [MSM_DP_CONTROLLER_2] = { .io_start = 0x0aea0000,
> .connector_type = DRM_MODE_CONNECTOR_eDP, .wide_bus_en = true },
>                  [MSM_DP_CONTROLLER_0] = { .io_start = 0x0ae90000,
> .connector_type = DRM_MODE_CONNECTOR_DisplayPort, .wide_bus_en = true },
>          },
>          .num_descs = 2,
> };
>
> 2)  DP always has index of 0 (dp->id = 0) and the first one to call
> msm_dp_modeset_init(). This make DP always place at head of bridge chain.

What does this mean? Are you talking about the list of bridges in drm
core, i.e. 'bridge_list'?

>
> At next patch eDP must be placed at head of bridge chain to fix eDP
> corruption issue. This is the purpose of this patch. I will revise the
> commit text.
>

Wouldn't that be "broken" again if we decided to change drm_bridge_add()
to add to the list head instead of list tail? Or if somehow
msm_dp_modeset_init() was called in a different order so that the DP
bridge was added before the eDP bridge?