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

[Archit Taneja <archit@ti.com>]

On Monday 10 December 2012 01:33 PM, Tomi Valkeinen wrote:
> On 2012-12-10 09:34, Archit Taneja wrote:
>> Hi,
>>
>> On Friday 07 December 2012 05:25 PM, Tomi Valkeinen wrote:
>>> Commit 5d89bcc341771d95e3a2996218e5949a6627f59e (OMAPDSS: remove initial
>>> display code from omapdss) moved setting up the initial overlay, overlay
>>> manager, output and display connections from omapdss to omapfb.
>>>
>>> However, currently omapfb only handles the connection related to the
>>> default display, which means that no overlay managers are connected to
>>> other displays.
>>>
>>> This patch changes omapfb to go through all dssdevs, and connect an
>>> overlay manager to them.
>>>
>>> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
>>> ---
>>>    drivers/video/omap2/omapfb/omapfb-main.c |   38
>>> +++++++++++++++++++-----------
>>>    1 file changed, 24 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/drivers/video/omap2/omapfb/omapfb-main.c
>>> b/drivers/video/omap2/omapfb/omapfb-main.c
>>> index 1df973e..24739fc 100644
>>> --- a/drivers/video/omap2/omapfb/omapfb-main.c
>>> +++ b/drivers/video/omap2/omapfb/omapfb-main.c
>>> @@ -2353,27 +2353,37 @@ static int omapfb_init_display(struct
>>> omapfb2_device *fbdev,
>>>    }
>>>
>>>    static int omapfb_init_connections(struct omapfb2_device *fbdev,
>>> -        struct omap_dss_device *dssdev)
>>> +        struct omap_dss_device *def_dssdev)
>>>    {
>>>        int i, r;
>>> -    struct omap_overlay_manager *mgr = NULL;
>>> +    struct omap_overlay_manager *mgr;
>>>
>>> -    for (i = 0; i < fbdev->num_managers; i++) {
>>> -        mgr = fbdev->managers[i];
>>> -
>>> -        if (dssdev->channel = mgr->id)
>>> -            break;
>>> +    if (!def_dssdev->output) {
>>> +        dev_err(fbdev->dev, "no output for the default display\n");
>>> +        return -EINVAL;
>>>        }
>>>
>>> -    if (i = fbdev->num_managers)
>>> -        return -ENODEV;
>>> +    for (i = 0; i < fbdev->num_displays; ++i) {
>>> +        struct omap_dss_device *dssdev = fbdev->displays[i].dssdev;
>>> +        struct omap_dss_output *out = dssdev->output;
>>>
>>> -    if (mgr->output)
>>> -        mgr->unset_output(mgr);
>>> +        mgr = omap_dss_get_overlay_manager(dssdev->channel);
>>
>> This dssdev->channel reference is something we would want to get rid of
>> eventually, right?
>
> Yes.
>
>> At the point omapfb_init_connections() is called, we would have all the
>> omap_dss_devices registered, right? So at this point, omapfb will have
>> an overall view of how the panels need to be connected to DSS.
>>
>> I think we can try to find a manager here for dssdev rather than using
>> dssdev->channel directly. The dssdev's output could connect to a few
>> managers. We would want to chose managers for each dssdev output in such
>> a way that all outputs have a manager. I guess there would be multiple
>> combinations for this, but it would be okay to pick any one of them.
>>
>> I think we would need some recursive or backtracking sort of approach to
>> get a desired combination. We can figure about how to make it work
>> later, but do you agree if this is a right way to get rid of
>> dssdev->channel?
>
> Yes, I think that's a sensible approach. The thing I don't like about it
> is that it requires omapfb/omapdrm to create the mgr-output connections.
> They shouldn't really be interested in that. All they want is a display
> device that works, and a way to get the mgr used for the display.
>
> Well, I think we can hide the implementation inside omapdss, so that
> omapfb/omapdrm will just need to call one omapdss func when they are
> started, which will connect the mgrs to outputs.
>
> A downside with setting up the mgr-output links at one go, when
> omapfb/omapdrm starts, is that it creates a strict load order
> restriction between omapdss, panels and omapfb/omapdrm. The drivers will
> need to be loaded in that order, or things won't work.

Yes, that's true.

>
> 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.

>
> 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.

>
> 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?


> 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

>
> 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?

Archit