Buildroot kernel creation

Rob Landley rob at landley.net
Sat Apr 1 18:12:12 UTC 2006


On Thursday 30 March 2006 6:06 am, Vincent Sanders wrote:
> > The advantage of miniconfig is that it's not version dependent.  You can
> > supply the same .config to the daily snapshot and life should generally
> > be good.  (It'll die with an error message if it tries to set a config
> > variable that doesn't exist, and if a new config entry shows up that you
> > actually _need_ to set it won't set it.  But to give you an example, my
> > UML mini.config from 2.6.12 still gives me a working 2.6.16-rc4 UML
> > vmlinux.
>
> sorry for the delay rob, been a bit busy (oh theres a surprise ;-)

Join the club, I'm catching up on the entire week's list posts at once. :)

> I like this idea a lot, did this get merged into 2.6.16? Might have to 
> improve the miniconfig.sh perf if thats a problem?

The problem is that Roman Zippel objected, and I'd touched his code.  I have a 
different approach that just touches the makefile and adds the shell script, 
and if I can get that working ok I'll probably submit that so he doesn't have 
standing to block it.

The miniconfig.sh is a hack, not much point in trying to improve it.  Ideally 
I'd want to convince the normal save path to spit out a miniconfig, but I 
haven't gone that deep into the logic yet.

> Even so I can rejig the kernel building stuff to use it which removes
> the need for a kernel config per release - much better. I am hoping to
> let the machine support packages select override configs anyway, this
> will adress the needs of the specific platforms like the current VIA
> solution.

In theory, in 2.6.15 and up you can use "allno.config" instead of 
"mini.config" and run "make allnoconfig".  The internal plumbing is there.  
The downsides of that are A) no obvious name, B) no error checking at all.  
(At won't notice if the file isn't there (typoed name), and it won't tell you 
if it has errors due to symbol name typos.  Actually it will spit out a 
warning to stderr but you have to redirect stdout to /dev/null to have any 
chance of seeing it.  And it won't exit with a detectable error code in 
either case.)

Rob
-- 
Never bet against the cheap plastic solution.



More information about the uClibc mailing list