-/* Use DMA to move data from on chip to external memory. While this is
- * required for only L1 instruction (it is not directly readable by the
- * core via data loads), it isn't a huge performance issue for other
- * regions (it's probably even faster than core load/stores). However,
- * the DMA engine does not have access to the L1 scratchpad, and we
- * cannot use DMA inside of the MMR space.
+/* Use DMA to move data from on chip to external memory. The L1 instruction
+ * regions can only be accessed via DMA, so if the address in question is in
+ * that region, make sure we attempt to DMA indirectly.