[Letux-kernel] Debugging 4.11-rc6 on GTA04

[H. Nikolaus Schaller]

Hi,

> Am 12.04.2017 um 11:39 schrieb Christ van Willegen <cvwillegen at gmail.com>:
> 
> On Tue, Apr 11, 2017 at 3:36 PM, H. Nikolaus Schaller <hns at goldelico.com> wrote:
> 
>> An idea could be to merge one after the other manually on top of the
>> minimal branch until it starts to fail. Then we know which feature
>> branch makes the problem.
> 
> This could be 'dangerous'!

Yes, you are right. One should do "normal" git bisect.

But in this case we can't. The reason is that some feature branches
are required to boot the device. And the others are piled on top.

Well, we could manually merge things in a different order to get
everything required into one good known commit and then add the others.

But it is not much different than rerunning my merge script with
different subsets of all features. My time consuming task is to
make clean+make and copy the result to the SD card and reboot.
I can't simply overwrite the uImage on the SD card without removing
it and placing it manually in a different SD reader.

So bugs that break boot are more difficult to fix.

> I remember testing 2 feature branches for a
> software company. Both worked out OK. When the features were merged,
> however, it turned out that the second would delete a file form the
> first, and the combination did not work as expected..

But still you know that when merging the second one as well it breaks
everything before. Then you can just look at what the second feature
does really change and you have a quite good chance to find out what
the combination is doing wrongly.

So the key issue is to find the offending feature or commit (which
is different from what Andreas has found in addition - which does not
seem to make problems for my setup).

What I have finally found after numerous hours trying different
combinations is that my problems seem to come from these two
feature branches:

extern/bno055
extern/mvduin/patch/tiler-fbdev

If I remove both I can merge all others and get device operation as
expected. So the musb effects are side-effects of something else.

Why the bno055 driver makes problems isn't obvious at all. Maybe it
is something with our iio-inputbridge.

Well and the tiler patches are known to not work well on OMAP3.
Ah, this reminds me that I have recently upgraded the tiler branch
for the Pyra and may have forgotten to apply some OMAP3 workarounds.

> 
> Great work on this, by the way!

Thanks for encouraging words (they have become and endangered species
these days)!

BR,
Nikolaus