As reported by Zhouyi Zhou, WRITE_ONCE() is not atomic
as expected when KASAN or KCSAN are compiled in.

Fix it by re-implementing it using inline assembly.

Reported-by: Zhouyi Zhou <zhouzhouyi@gmail.com>
Fixes: 077fc62b2b66 ("powerpc/irq: remove inline assembly in hard_irq_disable macro")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
---
 arch/powerpc/include/asm/hw_irq.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h
index 8a7b0b78a80e..83ab6668387b 100644
--- a/arch/powerpc/include/asm/hw_irq.h
+++ b/arch/powerpc/include/asm/hw_irq.h
@@ -285,7 +285,8 @@ static inline bool pmi_irq_pending(void)
 	flags = irq_soft_mask_set_return(IRQS_ALL_DISABLED);		\
 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;			\
 	if (!arch_irqs_disabled_flags(flags)) {				\
-		WRITE_ONCE(local_paca->saved_r1, current_stack_pointer);\
+		asm volatile("std%X0 %1,%0" : "=m" (local_paca->saved_r1) \
+					    : "r" (current_stack_pointer)); \
 		trace_hardirqs_off();					\
 	}								\
 } while(0)
-- 
2.37.1

On Tue, 23 Aug 2022 18:36:35 +0200, Christophe Leroy wrote:
> As reported by Zhouyi Zhou, WRITE_ONCE() is not atomic
> as expected when KASAN or KCSAN are compiled in.
> 
> Fix it by re-implementing it using inline assembly.
> 
> 

Applied to powerpc/fixes.

[1/1] powerpc: Fix hard_irq_disable() with sanitizer
      https://git.kernel.org/powerpc/c/814816d71e29934d0a76ee259b54c0b80c3b0e4a

cheers