Re: [PATCH 5/5] OMAPFB: connect ovl managers to all dssdevs

[Archit Taneja <archit@ti.com>]

On Monday 10 December 2012 03:37 PM, Tomi Valkeinen wrote:
> On 2012-12-10 11:54, Archit Taneja wrote:
>> On Monday 10 December 2012 01:33 PM, Tomi Valkeinen wrote:
>
>>> Another option would be to pass information about mgr-output links from
>>> the board files (or DT data) to the omapdss driver, so that omapdss
>>> could setup them at probe time. With this option omapfb/omapdrm doesn't
>>> need to care about this, and it doesn't create load order restriction.
>>> But mgr-output links are something that I'd really like to handle inside
>>> the drivers, not something that needs to be passed via platform data.
>>
>> This would definitely make things simpler, but if this parameter is put
>> in a panel's DT, it would become omap specific. We could add this info
>> to the DT corresponding to omapdss.
>
> Yes, I meant it should be omapdss platform data. Nothing related to panels.

I think this is the easiest way out. We can have one parameter per 
output in DT. If we do come up with a way to implement the 3rd option 
below, we can always ignore those DT fields(assuming our implementation 
to give the same result). So in this way, we would just be deprecating a 
DT field in the future, and calculating it ourselves.

>
>>> Third option, which is the best, but also something I have no idea how
>>> to implement, would be to create the mgr-output links dynamically when
>>> needed. The problem here is, of course, that a mgr could be allocated to
>>> an output, only to be later found out that that particular mgr is needed
>>> for another output.
>>>
>>> But this is something we could study a bit: can we create such mgr
>>> allocation system, that no matter what outputs the board uses, it'll
>>> just work.
>>
>> Yes, that would be quite useful. But I think we'll hit situations where
>> it is sort of impossible to prevent the above situation.
>>
>> When an output needs a manager. We could study the current state of the
>> system by splitting managers into 2 sets:
>>
>> A: managers which already have outputs connected to them
>> B: managers which don't have an output, but might get connected to one
>> in the future.
>>
>> managers in A are lost, and we can't detach them, we would need to at
>> least disable/reenable the panel with a new manager connected to the
>> output.
>>
>> we need to find one from B such that maximum outputs(or some other
>> weightage factor) will still be supported after this new link is made.
>>
>> The system will initially have all managers in B, but eventually
>> managers will move to A. We need to move one manager from B to A for
>> every mgr-output link.
>>
>> I guess I just described the problem in a more mathematical way, without
>> providing any solution :), but it does look like an optimisation problem.
>
> Well, optimization problem sounds like something that can always be
> solved. But in this case the driver may need to predict what outputs
> will be used, which is of course impossible.
>
>>> So, for example, on omap4, LCD2 mgr can be used for DPI, DSI2, DSI1, and
>>> RFBI. LCD1 can be used for RFBI and DSI1. If we know that DSI1 and RFBI
>>> cannot be used at the same time, we're free to give LCD1 to either one.
>>
>> But they can be used together, can't they? LCD1->DSI1 and LCD2->RFBI.
>> Are you creating this constraint by assuming what the board is like? Or
>> is this a constraint of OMAP DSS HW?
>
> I didn't check if they can be used together or not, I was just guessing.
> At least on OMAP3 DSI and RFBI shared the same pins, so they could not
> be used at the same time.
>
> Perhaps we should implement a mixed approach: the driver presumes
> certain things, like "if DSI is used, RFBI is not used", based on the
> knowledge of what kind of boards there currently are. This would allow
> us to manage the mgr->output connections in the driver for, say, 95% of
> the cases. Then we'd also have the platform data parameters for omapdss,
> which could be used in the weird cases.

Let's just have platform data parameters :)

>
>>> And if we know that DPI and DSI2 cannot be used at the same time, we're
>>> also free to give LCD2 to either one. And if that's the case, there are
>>> no conflicts.
>>
>> This is also possible at the same time: TV->DPI and LCD2->DSI2
>
> True. I was just again guessing. On OMAP3 DPI and DSI shared the same pins.
>
> Thinking about this, OMAP4 does have separate pins for DSI, doesn't it?
> So my guesses don't hold.
>
>>> I don't know if we can find such allocation for all current omaps, and I
>>> know it's slightly risky as the next omap could have limitations even if
>>> current ones do not.
>>
>> I don't understand the example so well, but I get your point of taking
>> advantage of such limitations.
>>
>>>
>>>> Also, we would need to do this for omapdrm separately using it's own
>>>> encoder/connector entities.
>>>
>>> Yep. That's also a negative side: both omapfb/omapdrm will need to
>>> implement the same stuff, even if neither of them are really interested
>>> in that stuff.
>>
>> Yes. I wonder if crtcs, encoders and connectors already have some sort
>> of helpers for this?
>
> Probably nothing that helps us, as this is OMAP HW restriction.

If we do use DT/platform data, would we need to parse it in omapdrm to 
establish drm entities? Or do we rely on omapdss to parse the DT data 
and give the links to omapdrm?

Archit