[Gta04-owner] We need an Android kernel

[Dr. H. Nikolaus Schaller]

Am 18.07.2013 um 21:49 schrieb Paul Kocialkowski:

> Le mercredi 17 juillet 2013 à 15:49 +0200, Dr. H. Nikolaus Schaller a
> écrit :
>>> It is far from perfect.
>> 
>> I didn't expect that it would be perfect...
> 
> I quickly looked at the diff from the android-3.4 kernel and mainline,
> and there are many changes there are there that 3.7 doesn't include. If
> we decide to stick with 3.7, we'll have to look at each of them
> carefully. Anyway the point is that there is a lot of work, and perhaps
> adding gta04-specific drivers and stuff on an android kernel would be
> easier (that is, if suspend/resume works).

I think we risk to have to integrate the gta04-specific drivers every now and then again.

> 
>> If nobody decides to go this way, uses and tests the kernel and others
>> are working to fixing things, it will never become stable.
> 
> I think we are still in the phase where we evaluate carefully our
> options and look at what is the easiest choice. I am still not convinced
> using 3.7 and adding Android stuff on top is the way to go. I'll also
> look at GNUtoo's work on 3.8 (which has complete Android support on
> AOSP's common kernel repo).

Christoph (Mair) has pointed me to this article where this evaluation has already
been done:

https://lwn.net/Articles/542466/

IMHO the key sentence (referring to 3.10) is: 

''With all of the staging code, John said, Android "just works" on a mainline kernel.''

This means that we should not waste any time working on older kernels
and rather build upon what others already have done. And with the 3.7 neil-plus
we are just three releases behind and should focus on a rebase to 3.10/3.11...

> 
>>> So the situation is the same as it was when we tested Neil's 3.7 kernel.
>>> However, it is true that suspend/resume works better than on the 3.0
>>> kernel.
>> 
>> That is IMHO an important point since I think nobody will be willing to
>> invest much into 3.0 or 3.4 suspend/resume issues if it works on 3.7 and
>> later.
> 
> That's right. Moreover, I'm not skilled to debug suspend/resume and it
> seems like such a huge task that I don't even want to try.

Me too... Therefore we won't get suspend/resume in a 3.4 kernel soon...

> 
>>> I think I'll give 3.4 a try, depending on how suspend/resume works
>>> there, it might be worth investigating that way instead.
>> 
>> Yes, for a comparison this will be very important to know. 
> 
> So I just spent the last 2 days trying to figure out why /dev remained
> empty on the android 3.4 kernel. Turns out it was the battery driver's
> fault (for some reason, it didn't work and made sysfs hang), I disabled
> it for now and will keep going by evaluating how well the rest works.
> 
>> Nevertheless I think it is more valuable investing time to fix the issues
>> that appear in the 3.7 kernel than working on (different) bugs of 3.4...
> 
> Again, that depends on the bugs and the complexity of the tasks. Merging
> all the Android changes seems more challenging than fixing separate
> gta04-specific drivers. I think we still need to clearly figure out what
> is the todo list for each kernel and decide then.
> 
>> Maybe you (and we as testers) can file the issues here:
>> 
>> http://projects.goldelico.com/p/gta04-kernel/issues/
>> 
>>> 
>>>> As soon as you are happy with it, please push to the repo so that I (we)
>>>> can sync and try ourselves...
>> 
>> Is it possible to "push" several different kernel trees to the Replicant repo?
>> I.e. that we can choose to repo-sync / compile the 3.0, 3.4 and 3.7 kernels?
> 
> It works like an usual git repo: you can push wherever you want, add
> other remotes, checkout their contents, etc. When a new commit is
> available from Replicant, it will be fetched by repo and may be checked
> out (replacing your current HEAD) if it's in the branch that is
> designated as the default one for replicant-4.0.
> 
> I will try to push as soon as things get interesting. Basically I fear
> that pushing too early will be a waste of time (for pusing) as well as
> disk space (everyone fetches the new branch with the next repo sync),

Ok, this should be avoided. I already filled 50 GByte just to compile replicant 2.3 and 4.0...

> so
> I'd rather do it only when it's worth it. On the other hand, I can also
> push somewhere else (like on my personal git.paulk.fr space): that
> shouldn't be a problem.

That would be fine!

Maybe to go forward you could use a suboptimal 3.4 kernel (w/o suspend/resume)
for Replicant 4.0 while we all try to move neil-plus to 3.10 (or 11 or 12) prepared for
let's say Replicant 4.3 as a more long-term goal.

So if you share your patches for the 3.7-neil-plus kernel we can start from there.

BR,
Nikolaus