Skip to contentSkip to conversation

Revert "kernel/kernel-arch: Explicitly mapping between, i386/x86_64 and x86 for kernel ARCH"

PATCH New Tom Zanussi 1 participant 1 message May 10, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "kernel/kernel-arch: Explicitly mapping between, i386/x86_64 and x86 for kernel ARCH"

This reverts commit a6f52930a68d8462e23486d51cdda715072dd752.

In addition to also causing the problem in [YOCTO #9579], this commit was reverted in krogoth and master but wasn't reverted in jethro but should be. The original revert message was:

This reverts commit 8d310b24927d0f348fb431895f0583733db2aad0.

That commit completely breaks KBUILD_DEFCONFIG because it relies on $ARCH to match between the target OE arch and the kernel subdirectory containing the defconfigs. In the kernel all defconfigs for everything x86-based (including x86_64) is stored in dir arch/x86/configs/

kernel-yocto.bbclass correctly searches for all the defconfigs inside ${S}/arch/${ARCH}/configs/${KBUILD_DEFCONFIG}

Commit 8d310b249 makes it search in wrong places and _only_ if you define TARGET_ARCH = "athlon" will it search x86 which is nonsensical.

The commit further adds an if clause to hack the mungled kernel arches back to their original values (ugh) in do_shared_workdir which is run after do compile, but of course the build breaks before that in do_kernel_metadata because of the KBUILD_DEFCONFIG mentioned above (so that hack is useless).

Please fix that corner case bug in another way which does not completely screw up the kernel arch mapping & defconfig logic. If 64bit configs are generated in the kernel for 32bit machines because the host is asked, then it it a bug in the kernel, it is of no use to hack around it in OE.

(From OE-Core rev: bc02a478a5d4a5de7b3943ed809d5c22711f5b1f)

---
 meta/classes/kernel-arch.bbclass |  4 +---
 meta/classes/kernel.bbclass      | 15 +++------------
 2 files changed, 4 insertions(+), 15 deletions(-)
meta/classes/kernel-arch.bbclass +1 -3
index d8b180e..3ed5986 100644
@@ -21,9 +21,7 @@ def map_kernel_arch(a, d):
 
     valid_archs = d.getVar('valid_archs', True).split()
 
-    if   re.match('i.86$', a):                  return 'i386'
-    elif re.match('x86.64$', a):                return 'x86_64'
-    elif re.match('athlon$', a):                return 'x86'
+    if   re.match('(i.86|athlon|x86.64)$', a):  return 'x86'
     elif re.match('armeb$', a):                 return 'arm'
     elif re.match('aarch64$', a):               return 'arm64'
     elif re.match('aarch64_be$', a):            return 'arm64'
meta/classes/kernel.bbclass +3 -12
index ee3e9a0..5e8b6cf 100644
@@ -309,18 +309,9 @@ do_shared_workdir () {
 		cp -fR include/generated/* $kerneldir/include/generated/
 	fi
 
-	# When ARCH is set to i386 or x86_64, we need to map ARCH to the real name of src
-	# dir (x86) under arch/ of kenrel tree, so that we can find correct source to copy.
-
-	if [ "${ARCH}" = "i386" ] || [ "${ARCH}" = "x86_64" ]; then
-		KERNEL_SRCARCH=x86
-	else
-		KERNEL_SRCARCH=${ARCH}
-	fi
-
-	if [ -d arch/${KERNEL_SRCARCH}/include/generated ]; then
-		mkdir -p $kerneldir/arch/${KERNEL_SRCARCH}/include/generated/
-		cp -fR arch/${KERNEL_SRCARCH}/include/generated/* $kerneldir/arch/${KERNEL_SRCARCH}/include/generated/
+	if [ -d arch/${ARCH}/include/generated ]; then
+		mkdir -p $kerneldir/arch/${ARCH}/include/generated/
+		cp -fR arch/${ARCH}/include/generated/* $kerneldir/arch/${ARCH}/include/generated/
 	fi
 }
 
Signature · 1 line
1.9.3