i386 install.sh script doesn't handle cross-compiling well

["Robert P. J. Day" <rpjday@xxxxxxxxxxxxxx>]

  (another bit of fallout from yesterday's cross-compiling adventure
... if this is a bit incoherent, it's late.)

  in cross-compiling the kernel for x86_64, i had the location of the
toolchain in my search path and all of my config and build commands
were of the form:

  $ make ARCH=x86_64 CROSS_COMPILE=x86_64- ...

for targets like

						... defconfig
						... menuconfig
						... bzImage
						... modules

so far, so good.

  then, even though it was certainly redundant, i left the ARCH= and
CROSS_COMPILE= make variables there when i did:

  # make ... modules_install
  # make ... install

  first, i'm guessing that, regardless of architecture or
cross-compile, you can always run simply "make modules_install",
correct?  is there anything about installing modules that needs to
know about architecture or cross compiling?  in any event, having
those variables there didn't seem to have any effect, but i'm assuming
they have no value for this target, right?

  on the other hand, it *did* have an effect on doing the kernel
install.  running

  # make ARCH=x86_64 CROSS_COMPILE=x86_64- install

invoked the script arch/x86_64/boot/install.sh, which consists
entirely of:

  #!/bin/sh
  . $srctree/arch/i386/boot/install.sh

ok, so off we go to the i386 install script, which ends with the
following:
======================================================================
...
# User may have a custom install script

if [ -x ~/bin/${CROSS_COMPILE}installkernel ]; then exec ~/bin/${CROSS_COMPILE}installkernel "$@"; fi
if [ -x /sbin/${CROSS_COMPILE}installkernel ]; then exec /sbin/${CROSS_COMPILE}installkernel "$@"; fi

# Default install - same as make zlilo

if [ -f $4/vmlinuz ]; then
        mv $4/vmlinuz $4/vmlinuz.old
fi

if [ -f $4/System.map ]; then
        mv $4/System.map $4/System.old
fi

cat $2 > $4/vmlinuz
cp $3 $4/System.map

if [ -x /sbin/lilo ]; then /sbin/lilo; else /etc/lilo/install; fi
=====================================================================

  so if you're doing a regular (i386) install, CROSS_COMPILE isn't set
and you call the regular /sbin/installkernel.

  OTOH, if you're doing an x86_64 install and you accidentally leave
CROSS_COMPILE in the "make" line, that script will obviously fail to
find the appropriate (custom) version of "installkernel", then
continue down until it tries to run "lilo".  ugh.  and that's what
happened when i ran

  # make ... CROSS_COMPILE=x86_64- install

and eventually got the error message:

  ... /etc/lilo/install:  No such file or directory.

it just seems sloppy to have had that script eventually work its way
down to trying to run lilo on this system, when lilo is not even
installed.

  thoughts?

rday

-- 
========================================================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://fsdev.net/wiki/index.php?title=Main_Page
========================================================================

--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ