[PATCH kernel] powerpc/powernv: Fix clobbered MSR

If CONFIG_DEBUG_INFO_SPLIT is not set but CONFIG_DEBUG_INFO is,
the kernel makefile just adds "-g" and the scripts/gcc-goto.sh test for
the "asm goto ("" :::: entry)" support succedes and adds
-DCC_HAVE_ASM_GOTO to KBUILD_CFLAGS/KBUILD_AFLAGS. This effectively
makes OPAL_BRANCH() a noop (or something similar).

With CONFIG_DEBUG_INFO_SPLIT=y, the makefile adds "-gsplit-dwarf" which
somehow makes the scripts/gcc-goto.sh test fail and not have
CC_HAVE_ASM_GOTO defined so the alternative OPAL_BRANCH() is used
and this particular chunk clobbers r12 where the parenting code -
OPAL_CALL() - stores MSR; as the result, the kernel oops'es right after
early_setup() because of broken MSR.

This replaces r12 with r11 which is overwritten right after
OPAL_BRANCH(opal_tracepoint_entry) anyway.

I used gcc 5.4.1 20161205 built from sha1 ffadbf3ae29.

Fixes: ab9bad0ead9a ("powerpc/powernv: Remove separate entry for OPAL real mode calls")
Suggested-by: Paul Mackerras <redacted>
Signed-off-by: Alexey Kardashevskiy <redacted>
---
 arch/powerpc/platforms/powernv/opal-wrappers.S | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/platforms/powernv/opal-wrappers.S b/arch/powerpc/platforms/powernv/opal-wrappers.S
index 6693f75e93d1..da8a0f7a035c 100644
--- a/arch/powerpc/platforms/powernv/opal-wrappers.S
+++ b/arch/powerpc/platforms/powernv/opal-wrappers.S

@@ -39,8 +39,8 @@ opal_tracepoint_refcount:
 BEGIN_FTR_SECTION;						\
 	b	1f;						\
 END_FTR_SECTION(0, 1);						\
-	ld	r12,opal_tracepoint_refcount@toc(r2);		\
-	cmpdi	r12,0;						\
+	ld	r11,opal_tracepoint_refcount@toc(r2);		\
+	cmpdi	r11,0;						\
 	bne-	LABEL;						\
 1:
 

-- 
2.11.0

On Thu, 2017-03-02 at 17:41 +1100, Alexey Kardashevskiy wrote:

With CONFIG_DEBUG_INFO_SPLIT=y, the makefile adds "-gsplit-dwarf" which
somehow makes the scripts/gcc-goto.sh test fail and not have
CC_HAVE_ASM_GOTO defined so the alternative OPAL_BRANCH() is used

This needs to be fixed too though. That alternative path has much more
overhead accross the board. Losing "asm goto" because of a broken test
is nasty.

Cheers,
Ben.

On 03/03/17 08:37, Benjamin Herrenschmidt wrote:

On Thu, 2017-03-02 at 17:41 +1100, Alexey Kardashevskiy wrote:

quoted

With CONFIG_DEBUG_INFO_SPLIT=y, the makefile adds "-gsplit-dwarf" which
somehow makes the scripts/gcc-goto.sh test fail and not have
CC_HAVE_ASM_GOTO defined so the alternative OPAL_BRANCH() is used

This needs to be fixed too though. That alternative path has much more
overhead accross the board. Losing "asm goto" because of a broken test
is nasty.

I just posted "[RFC PATCH kernel] gcc-goto.sh: Allow to pass with
CONFIG_DEBUG_INFO_SPLIT=y" to Jason + cc: Michael, will see how it goes.



-- 
Alexey

On Thu, 2017-03-02 at 06:41:24 UTC, Alexey Kardashevskiy wrote:

If CONFIG_DEBUG_INFO_SPLIT is not set but CONFIG_DEBUG_INFO is,
the kernel makefile just adds "-g" and the scripts/gcc-goto.sh test for
the "asm goto ("" :::: entry)" support succedes and adds
-DCC_HAVE_ASM_GOTO to KBUILD_CFLAGS/KBUILD_AFLAGS. This effectively
makes OPAL_BRANCH() a noop (or something similar).

With CONFIG_DEBUG_INFO_SPLIT=y, the makefile adds "-gsplit-dwarf" which
somehow makes the scripts/gcc-goto.sh test fail and not have
CC_HAVE_ASM_GOTO defined so the alternative OPAL_BRANCH() is used
and this particular chunk clobbers r12 where the parenting code -
OPAL_CALL() - stores MSR; as the result, the kernel oops'es right after
early_setup() because of broken MSR.

This replaces r12 with r11 which is overwritten right after
OPAL_BRANCH(opal_tracepoint_entry) anyway.

I used gcc 5.4.1 20161205 built from sha1 ffadbf3ae29.

Fixes: ab9bad0ead9a ("powerpc/powernv: Remove separate entry for OPAL real mode calls")
Suggested-by: Paul Mackerras <redacted>
Signed-off-by: Alexey Kardashevskiy <redacted>

Applied to powerpc fixes, thanks.

https://git.kernel.org/powerpc/c/2a9c4f40ab2c281f95d41577ff0f7f

cheers