[meta-ti] building Yocto for Pandaboard

[jfabernathy]

On 02/09/2012 10:33 AM, Richard Purdie wrote:
> On Thu, 2012-02-09 at 14:54 +0100, Koen Kooi wrote:
>> Op 9 feb. 2012, om 14:11 heeft jfabernathy het volgende geschreven:
>>> On 02/09/2012 08:01 AM, Maupin, Chase wrote:
>>>>> Wow!  Sorry I jumped into a mailing list I obviously don't
>>>>> understand or belong in.  I apologize if I offended.  I did read
>>>>> the README, but it didn't make a bit of sense to me because it
>>>>> talked about angstrom, which I don't know anything about and
>>>>> wondered what that had to do with yocto.  My current Yocto
>>>>> knowledge is based on the meta-intel layer, which doesn't mention
>>>>> angstrom.  It sounds like the hint/trick that Gary mentioned will
>>>>> make bitbake build with just yocto, which is what I want.  My goal
>>>>> is more of a proof of concept.  I'd like to prove if you could take
>>>>> the same image recipe and move it from Pandaboard to Atom and vice
>>>>> versa.  That way a developer could pick the hardware platform based
>>>>> on the performance, features, and cost.  The software effort should
>>>>> be minimal to move if the Yocto concept works as advertised.
>>>> Jim, you are welcome here.  As was mentioned before there have been
>> a lot of discussions about this layer.  One of the goals for the
>> meta-ti layer will be to work with just oe-core and yocto and not
>> require meta-angstrom.  We are moving that direction and the use case
>> you are trying and your experiences with it are important.  Thanks for
>> taking the time to give this a shot.
>>> Thanks, glad the proof of concept falls into the groups thinking as
>> well.  I will monitor for progress on the use of just the Yocto Linux
>> for this POC.
>>
>> Yocto or Poky? Because if you want yocto, angstrom very much *is*
>> yocto. If you want Poky, that's something different.
> I think there are a few things need to get cleared up here as this is
> likely confusing for people.
>
> The Yocto Project is the overall project which is working on improving
> embedded Linux for people in whatever areas those improvements are
> needed in. This involves things like eglibc, creating standard formats
> for hardware definitions (BSPs) and has also involved some work on build
> system which are derived from OpenEmbedded.
>
> The Yocto Project uses Poky as its reference build system which is a
> combination of Bitbake, OE-Core, a load of documentation and a very
> small shim called meta-yocto. Its purpose is to provide a known good
> base system people can build on top of by adding layers.
>
> Angstrom is similar to Poky in that its a mix of components allowing
> people to build things. It includes Bitbake and OE-Core and some other
> pieces. It has an equivalent to meta-yocto with different functionality.
> I'll let Koen as one of its maintainers describe what its purpose is.
> One difference is that is has a specific set of hardware targets and
> provides prebuilt package feeds which Poky does not (nor does the Yocto
> Project).
>
> meta-ti is intended as a hardware support layer for various TI products.
>
> The trouble is that some of these components have been around for a
> while and it hasn't always been possible to neatly layer things. In
> particular, meta-ti historically has a dependency on Angstrom.
>
> Work is underway to remove that dependency so meta-ti can operate freely
> with both Poky and Angstrom, or just with OE-Core and Bitbake as
> separate pieces. There is no architectural problem with this and it is
> the intention of everyone involved. There is the question of sorting
> some of the mechanics and making the change whilst not causing problems
> for existing users. I'm really happy to see progress on this!
>
> So in answer to Jim's question, you will be able to use meta-ti with
> Poky directly but right now, there are some details being worked through
> and it doesn't quite work. Perhaps we could mention that in the meta-ti
> README so people can see what the plan is as well as the current
> situation?
>
> Cheers,
>
> Richard
>
When you look at Yocto project from a marketing point of view, which is 
not something the open source community usually concentrates on for 
obvious reasons, it would be a very positive message to be able to talk 
about, and demonstrate with examples of taking the same project and 
moving it from one architecture to another with minimal effort, as long 
as both architectures supported the key features like media playback 
acceleration, 3D, sound, HDMI, etc.

While hardware companies tend to cringe when you make it too easy to 
move back and forth, the reality is, if you have a high volume product 
you can't justify or afford to throw excess performance, thermals, 
power, and cost where it is not needed.  For example a cheap digital 
signage application that just throws advertisement, or menu data on the 
screen could be done with an ARM processor like OMP3/4, but if you had a 
higher end product that needed to do extensive video analytics for 
customer profiling and advertising effectiveness, you're going to want a 
Core i5/i7.  It would be of great benefit to take the Yocto project that 
did the digital sign on the ARM and quickly get it up and running on the 
Core i5 and then concentrate on the analytics application.

I think this portability is a better sell, if it's all Yocto.  We all 
know that really smart people can move around from different Linux 
versions and development methods, but it's an easier sale to a customer 
management when you say it can all be done on Yocto and that it's easily 
demonstrated.

JIm A

>
>
>