Excerpts from Peter Zijlstra's message of July 23, 2020 9:40 pm:

On Thu, Jul 23, 2020 at 08:56:14PM +1000, Nicholas Piggin wrote:

quoted

diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h
index 3a0db7b0b46e..35060be09073 100644
--- a/arch/powerpc/include/asm/hw_irq.h
+++ b/arch/powerpc/include/asm/hw_irq.h

@@ -200,17 +200,14 @@ static inline bool arch_irqs_disabled(void)
 #define powerpc_local_irq_pmu_save(flags)			\
 	 do {							\
 		raw_local_irq_pmu_save(flags);			\
-		trace_hardirqs_off();				\
+		if (!raw_irqs_disabled_flags(flags))		\
+			trace_hardirqs_off();			\
 	} while(0)
 #define powerpc_local_irq_pmu_restore(flags)			\
 	do {							\
-		if (raw_irqs_disabled_flags(flags)) {		\
-			raw_local_irq_pmu_restore(flags);	\
-			trace_hardirqs_off();			\
-		} else {					\
+		if (!raw_irqs_disabled_flags(flags))		\
 			trace_hardirqs_on();			\
-			raw_local_irq_pmu_restore(flags);	\
-		}						\
+		raw_local_irq_pmu_restore(flags);		\
 	} while(0)

You shouldn't be calling lockdep from NMI context!

After this patch it doesn't.

trace_hardirqs_on/off implementation appears to expect to be called in NMI 
context though, for some reason.

That is, I recently
added suport for that on x86:

  https://lkml.kernel.org/r/20200623083721.155449112@infradead.org
  https://lkml.kernel.org/r/20200623083721.216740948@infradead.org

But you need to be very careful on how you order things, as you can see
the above relies on preempt_count() already having been incremented with
NMI_MASK.

Hmm. My patch seems simpler.

I don't know this stuff very well, I don't really understand what your patch 
enables for x86 but at least it shouldn't be incompatible with this one 
AFAIKS.

Thanks,
Nick