Re: [PATCH/RFC] kbuild: use INSTALLKERNEL to select customized installkernel script

[Sam Ravnborg <sam@ravnborg.org>]

On Mon, Jul 20, 2009 at 06:22:43PM -0700, H. Peter Anvin wrote:
> Russell King - ARM Linux wrote:
> > 
> > That's a good thing in some ways and a bad thing in others.  Having
> > it guarantee to fail if you try installing an ARM kernel onto an
> > x86 machine is definitely a good thing.  That said, it's not something
> > I've ever accidentally done.
> > 
> 
> I guess we could have it error out if you are building a non-native
> architecture but don't set CROSS_COMPILE.  Either that or we could
> default $INSTALLKERNEL to $(CROSS_COMPILE)installkernel, which would be
> backwards compatible.
I had googled a bit to try to analyse how much use
this feature has seen.
And I only managed to find the original patch implementing this.

None of the books I found mention this in their
install kernel chapets either.

So the conclusion - it is not widely used.

Russell is right that one implicit use is that "make CROSS_COMPILE=foo install"
would fail. And with this patch it would succeed.
Trying to error out is we are doing a non-native build is not the
right answer. People may like to have a custom script that can
ftp the new kernel somewhere or stuff like that. 

All that combined with the fact that using the binutils/gcc
CROSS_COMPILE prefix is the wrong solution anyway made me
conclude that we should fix this rather than implement it
in a different way.

That combined with the fact that CROSS_COMPILE has two
primary usages:

   - binutils/gcc prefix when building for another architecture

   - distcc/ccache

The latter does not fit well with the installkernel use.

I briefly looked into creating another way to specify
distcc / ccache.
But the current way to do things works well - so I dropped it again.

	Sam