Re: [PATCH V7 11/12] Documentation: bridge: Add documentation for ps8622 DT properties

[Andrzej Hajda <a.hajda@samsung.com>]

On 09/23/2014 01:52 PM, Laurent Pinchart wrote:
> On Tuesday 23 September 2014 13:47:40 Andrzej Hajda wrote:
>> On 09/23/2014 01:23 PM, Laurent Pinchart wrote:
>>> On Tuesday 23 September 2014 13:18:30 Andrzej Hajda wrote:
>>>> On 09/23/2014 01:10 PM, Laurent Pinchart wrote:
>>>>> On Tuesday 23 September 2014 12:02:45 Andrzej Hajda wrote:
>>>>>> On 09/23/2014 11:30 AM, Tomi Valkeinen wrote:
>>>>>>> On 23/09/14 09:21, Thierry Reding wrote:
>>>>>>>>> Well, I can write almost any kind of bindings, and then evidently my
>>>>>>>>> device would work. For me, on my board.
>>>>>>>>
>>>>>>>> Well, that's the whole problem with DT. For many devices we only have
>>>>>>>> a single setup to test against. And even when we have several they
>>>>>>>> often are derived from each other. But the alternative would be to
>>>>>>>> defer (possibly indefinitely) merging support for a device until a
>>>>>>>> second, wildly different setup shows up. That's completely
>>>>>>>> unreasonable and we need to start somewhere.
>>>>>>>
>>>>>>> Yes, but in this case we know of existing boards that have complex
>>>>>>> setups. It's not theoretical.
>>>>>>>
>>>>>>> I'm not saying we should stop everything until we have a 100% solution
>>>>>>> for the rare complex cases. But we should keep them in mind and, when
>>>>>>> possible, solve problems in a way that will work for the complex cases
>>>>>>> also.
>>>>>>>
>>>>>>>>> I guess non-video devices haven't had need for those. I have had
>>>>>>>>> lots of boards with video setup that cannot be represented with
>>>>>>>>> simple phandles. I'm not sure if I have just been unlucky or what,
>>>>>>>>> but my understand is that other people have encountered such boards
>>>>>>>>> also. Usually the problems encountered there have been circumvented
>>>>>>>>> with some hacky video driver for that specific board, or maybe a
>>>>>>>>> static configuration handled by the boot loader.
>>>>>>>>
>>>>>>>> I have yet to encounter such a setup. Can you point me at a DTS for
>>>>>>>> one such setup? I do remember a couple of hypothetical cases being
>>>>>>>> discussed at one time or another, but I haven't seen any actual DTS
>>>>>>>> content where this was needed.
>>>>>>>
>>>>>>> No, I can't point to them as they are not in the mainline (at least
>>>>>>> the ones I've been working on), for obvious reasons.
>>>>>>>
>>>>>>> With a quick glance, I have the following devices in my cabinet that
>>>>>>> have more complex setups: OMAP 4430 SDP, BeagleBoneBlack + LCD, AM43xx
>>>>>>> EVM. Many Nokia devices used to have such setups, usually so that the
>>>>>>> LCD and tv-out were connected to the same video source.
>>>>>>>
>>>>>>>>> Do we have a standard way of representing the video pipeline with
>>>>>>>>> simple phandles? Or does everyone just do their own version? If
>>>>>>>>> there's no standard way, it sounds it'll be a mess to support in the
>>>>>>>>> future.
>>>>>>>>
>>>>>>>> It doesn't matter all that much whether the representation is
>>>>>>>> standard.
>>>>>>>
>>>>>>> Again, I disagree.
>>>>>>>
>>>>>>>> phandles should simply point to the next element in the pipeline and
>>>>>>>> the OS abstractions should be good enough to handle the details about
>>>>>>>> how to chain the elements.
>>>>>>>
>>>>>>> I, on the other hand, would rather see the links the other way around.
>>>>>>> Panel having a link to the video source, etc.
>>>>>>>
>>>>>>> The video graphs have two-way links, which of course is the safest
>>>>>>> options, but also more verbose and redundant.
>>>>>>>
>>>>>>> When this was discussed earlier, it was unclear which way the links
>>>>>>> should be. It's true that only links to one direction are strictly
>>>>>>> needed, but the question raised was that if in the drivers we end up
>>>>>>> always going the links the other way, the performance penalty may be
>>>>>>> somewhat big. (If I recall right).
>>>>>>
>>>>>> I do not see why performance may drop significantly?
>>>>>> If the link is one-way it should probably work as below:
>>>>>> - the destination registers itself in some framework,
>>>>>> - the source looks for the destination in this framework using phandle,
>>>>>> - the source starts to communicate with the destination - since now
>>>>>> full two way link can be established dynamically.
>>>>>>
>>>>>> Where do you see here big performance penalty?
>>>>>
>>>>> The performance-related problems arise when you need to locate the
>>>>> remote device in the direction opposite to the phandle link direction.
>>>>> Traversing a link forward just involves a phandle lookup, but traversing
>>>>> it backwards isn't possible the same way.
>>>>
>>>> But you do not need to traverse backwards. You just wait when the source
>>>> start to communicate with the destination, at this moment destination can
>>>> build back-link dynamically.
>>>
>>> Your driver might not need it today for your use cases, but can you be
>>> certain that no driver on any OS would need to ?
>>
>> I have just showed how to create back-link dynamically if we have only
>> forward-link in DT.
>> Ie it is a trivial 'proof' that the direction is not so important.
>> So I do not understand why do you pose such question?
>>
>>> This becomes an issue even on Linux when considering video-related devices
>>> that can be part of either a capture pipeline or a display pipeline. If
>>> the link always goes in the data flow direction, then it will be easy to
>>> locate the downstream device (bridge or panel) from the display controller
>>> driver, but it would be much more difficult to locate the same device from
>>> a camera driver as all of a sudden the device would become an upstream
>>> device.
>>
>> Why?
>>
>> If you have graph:
>> sensor --> camera
>>
>> Then camera register itself in some framework as a destination device
>> and sensor looks in this framework for the device identified by remote
>> endpoint.
>> Then sensor tells camera it is connected to it and voila.
> 
> Except that both kernelspace and userspace deal with cameras the other way 
> around, the master device is the camera receiver, not the camera sensor. DRM 
> is architected the same way, with the component that performs DMA operations 
> being the master device.

But the link direction do not determines who should be the master
device. It just determines who should perform initial handshake.

Andrzej