+
+       *cs++ = MI_BATCH_BUFFER_START_GEN8 | MI_BATCH_SECOND_LEVEL;

You are not a second level batch. You are calling from the ring to a
global address of _0_.

+       *cs++ = 0;

low 32bits = 0

+       *cs++ = i915_ggtt_offset(stream->noa_reprogram_vma);

high 32bits = address you wanted. (order reversed)