Unfortunately, we still have a mess wrt dual-bitness arches.
For sparc, there's arch/sparc, with ARCH=sparc being "whatever the toolchain is",
ARCH=sparc32 being "give me a 32-bit kernel", and ARCH=sparc64 being "give me
a 64-bit kernel".
You do, seemingly, need to actually pass ARCH=sparc64 to guarantee correct results,
as various kernel Makefiles check for the ARCH value, even though ARCH=sparc will
pop out a 64-bit kernel.
(I originally updated toolchain-funcs.eclass's tc-ninja_magic_to_arch, but we later found that doesn't look particularly sturdy.
Per at the very least 5ba800962a80d4158b73fb91a7779df7b770c750 ("kbuild: update ARCH alias info for sparc")
upstream, although it seems to go back a fair bit earlier, ARCH=sparc when building
the kernel means "sparc64" if on a sparc64 host.)
Signed-off-by: Sam James <
[email protected]>
---
eclass/kernel-build.eclass | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/eclass/kernel-build.eclass b/eclass/kernel-build.eclass
index ff615a02d0ebf..276a08a104e0e 100644
--- a/eclass/kernel-build.eclass
+++ b/eclass/kernel-build.eclass
@@ -257,6 +257,14 @@ kernel-build_src_install() {
local kern_arch=$(tc-arch-kernel)
local dir_ver=${PV}${KV_LOCALVERSION}
local kernel_dir=/usr/src/linux-${dir_ver}
+
+ if use sparc ; then
+ # We don't want tc-arch-kernel's sparc64, even though we do
+ # need to pass ARCH=sparc64 to the build system. It's a quasi-alias
+ # in Kbuild.
+ kern_arch=sparc
+ fi
+
dodir "${kernel_dir}/arch/${kern_arch}"
mv include scripts "${ED}${kernel_dir}/" || die
mv "arch/${kern_arch}/include" \
--
2.41.0
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)