Cool, I still recognize the Instruction union as he based it on mine (libriscv). It's interesting to see how it has mutated across two implementations. Anyway,

Cool, I still recognize the Instruction union as he based it on mine (libriscv). It's interesting to see how it has mutated across two implementations. Anyway, this is overflowing with cool ideas. I do see one minor nitpick though. The inlined host call needs to also use an "m" operand, otherwise the compiler could optimize out the actual data (making it a dead-store). asm volatile("scall" : "+r"(a0) : "r"(a1), "r"(a7)); to asm volatile("scall" : "+r"(a0) : "r"(a1), "m"(*a1), "r"(a7)); Usually, we want to cast a1 here to a fixed or variable-length array in order to inform the compiler of the size of the data, but here the size is known. I believe it's enough with a deref. If not, cast it like so: *(uintptr_t(*)[13]) a1