On Wed, Nov 20, 2024 at 08:13:38AM -0700, Nathan Chancellor wrote:
Hi Sasha,
On Wed, Nov 20, 2024 at 09:06:35AM -0500, Sasha Levin wrote:
From: Linus Walleij <linus.walleij@xxxxxxxxxx>
[ Upstream commit 4aea16b7cfb76bd3361858ceee6893ef5c9b5570 ]
When enabling expert mode CONFIG_EXPERT and using that power
user mode to disable the branch prediction hardening
!CONFIG_HARDEN_BRANCH_PREDICTOR, the assembly linker
in CLANG notices that some assembly in proc-v7.S does
not have corresponding C call sites, i.e. the prototypes
in proc-v7-bugs.c are enclosed in ifdef
CONFIG_HARDEN_BRANCH_PREDICTOR so this assembly:
SYM_TYPED_FUNC_START(cpu_v7_smc_switch_mm)
SYM_TYPED_FUNC_START(cpu_v7_hvc_switch_mm)
Results in:
ld.lld: error: undefined symbol: __kcfi_typeid_cpu_v7_smc_switch_mm
>>> referenced by proc-v7.S:94 (.../arch/arm/mm/proc-v7.S:94)
>>> arch/arm/mm/proc-v7.o:(.text+0x108) in archive vmlinux.a
ld.lld: error: undefined symbol: __kcfi_typeid_cpu_v7_hvc_switch_mm
>>> referenced by proc-v7.S:105 (.../arch/arm/mm/proc-v7.S:105)
>>> arch/arm/mm/proc-v7.o:(.text+0x124) in archive vmlinux.a
Fix this by adding an additional requirement that
CONFIG_HARDEN_BRANCH_PREDICTOR has to be enabled to compile
these assembly calls.
Closes: https://lore.kernel.org/oe-kbuild-all/202411041456.ZsoEiD7T-lkp@xxxxxxxxx/
Reported-by: kernel test robot <lkp@xxxxxxxxx>
Reviewed-by: Nathan Chancellor <nathan@xxxxxxxxxx>
Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx>
Signed-off-by: Russell King (Oracle) <rmk+kernel@xxxxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
arch/arm/mm/proc-v7.S | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/arm/mm/proc-v7.S b/arch/arm/mm/proc-v7.S
index 193c7aeb67039..bea11f9bfe856 100644
--- a/arch/arm/mm/proc-v7.S
+++ b/arch/arm/mm/proc-v7.S
@@ -93,7 +93,7 @@ ENTRY(cpu_v7_dcache_clean_area)
ret lr
ENDPROC(cpu_v7_dcache_clean_area)
-#ifdef CONFIG_ARM_PSCI
+#if defined(CONFIG_ARM_PSCI) && defined(CONFIG_HARDEN_BRANCH_PREDICTOR)
.arch_extension sec
ENTRY(cpu_v7_smc_switch_mm)
This patch is unnecessary in branches prior to 6.10 (when ARM started
supporting kCFI) because SYM_TYPED_FUNC_START() is not used here. I
would just drop it for 6.6 and earlier.
Ack, will do.
--
Thanks,
Sasha
