Hi Michael,

quoted

+static void __init fixup_cpu_features(void)
+{
+	unsigned long version = mfspr(SPRN_PVR);
+
+	if ((version & 0xffffffff) == 0x00800100)
+		cur_cpu_spec->cpu_features |= CPU_FTR_POWER10_DD1;
+}
+
  static int __init early_init_dt_scan_cpus(unsigned long node,
  					  const char *uname, int depth,
  					  void *data)

@@ -378,6 +386,7 @@ static int __init early_init_dt_scan_cpus(unsigned long node,
  
  		check_cpu_feature_properties(node);
  		check_cpu_pa_features(node);
+		fixup_cpu_features();
  	}

This is not the way we normally do CPU features.

In the past we have always added a raw entry in cputable.c, see eg. the
Power9 DD 2.0, 2.1 entries.

True. But that won't work PowerVM guests because kernel overwrites "raw"
mode cpu_features with "architected" mode cpu_features.

Doing it here is not really safe, if you're running with an architected
PVR (via cpu-version property), you can't set the DD1 feature, because
you might be migrated to a future CPU that doesn't have the DD1 quirks.

Okay.. I suppose, that mean kernel need to check real PVR every time when it
encounters spurious exception. i.e. instead of using

   if (cpu_has_feature(CPU_FTR_POWER10_DD1) && ...)

it need to use

   if (mfspr(SPRN_PVR) == 0x800100 && ...)

in patch 2. Right?

Thanks,
Ravi