On 1/28/15 11:16 , Wolfgang Denk wrote:
Dear Richard,
In message <[email protected]> you wrote:
I should have read this thread to the end. You are right again, and
although my previous quick&dirty patch is good enough to build a
Linux kernel, this one needs fixing, too.
It's not good enough for some kernels. As I mentioned earlier, at least
the port I'm trying to use, (looks like all powerpc's, actually),
overwrite those variables at the Makefile level so the env versions are
never seen.
I tested current mainline with two different configurations (using
mpc5200_defconfig and mpc85xx_defconfig). Both build fine with just
sourcing the env file, un-setting LDFLAGS, and running "make" without
any specific command line options, i. e. just
-> make mpc85xx_defconfig
-> make -s -j8 uImage
In which way is your kernel different? where does it throw errors?
1. produces scary /warnings /on -mcpu, which is apparently now obsolete.
2. fails with an error on linking vmlinux because it can't find "libgcc.a".
Further investigation showed that the value "libgcc.a" was coming from
"mumble-gcc -print-libgcc-file-name" which produces different values on
powerpc-4xx, (full absolute path name), and powerpc-e500v2, (just
"libgcc.a"), as reported earlier. Passing -sysroot gets the
powerpc-e500v2 toolchain to produce a useful result from
-print-libgcc-file-name.
I agree that if -print-libgcc-file-name behaved similarly on
powerpc-e500v2 to the way it behaves on powerpc-4xx, then my command
line CC passing shouldn't be necessary. But for now, with eldk-5.6,
passing CC seems to be a reasonable workaround.
I'm not sourcing any env file. But I /am/ setting PATH in the
environment, and CROSS_COMPILE and CC on the make command line.
--rich
_______________________________________________
eldk mailing list
[email protected]
http://lists.denx.de/mailman/listinfo/eldk
