Re: Changing a machine def and rebuilding tool chain?

[GNUtoo <GNUtoo@no-log.org>]

On Thu, 2009-05-21 at 17:57 +0200, GNUtoo wrote:
> On Thu, 2009-05-21 at 09:22 -0500, Jeff Rush wrote:
> > I'm working on a machine definition for the old TuxScreen hardware as I
> > haven't seen one already defined in the OE metadata.
> > 
> > Initially I had a:
> > 
> >    TARGET_CC_ARCH = "-march=armv4t -mtune=strongarm"
> > 
> > where the TuxScreen CPU is a StrongARM 1100.  After building a
> > statically-lined "helloworld" program however, it gives me an instruction
> > error when I try to run it on the target hardware.
> > 
> > Figuring it was my use of "armv4t" I changed it to just "-march=armv4
> > -mtune=strongarm".  Doing a:
> > 
> >   $ bitbake -c clean helloworld
> >   $ bitbake helloworld
> > 
> > produces the exact same binary, so obviously I need to rebuild something
> > further upstream in my toolchain.  Does the BitBake system notice that a
> > machine definition has changed, or that a TARGET_CC_ARCH has changed and
> > rebuild the necessary pieces?  Or do I have to blow it all away and rebuild
> > everything whenever something as fundamental as the TARGET_CC_ARCH changes?
> > 
> > Thanks,
> > 
> > -Jeff
> according to http://www.linuxdevices.com/products/PD5824679728.html you
> have a 2.4 kernel.
> Did you try with a 2.4 kenrel, if so the problem could be the arm ABI
> (binaries are produced for EABI and the kenrel support OABI)
> here are more infos on arm ABI:
> http://wiki.debian.org/ArmEabiPort
> 
> I've attached my config for a 2.4 kenrel and oabi in case you have to
> fight against setting openembedded to cross compile to OABI
> 
> Unfortunately some variables were printed after the parsing aren't
> printed anymore...that could help a lot for debuging theses
> problems...maybe you will have to use peek in bitbake -i then...
or to use bitbake -D as I was told on IRC

Denis.