Fix - JIT ExceededMaxInstructions and ExecutionOverrun

[Lichtso]

Thanks to @ravyu-jump, @ripatel-fd and @topointon-jump for finding this.

[LucasSte]

Why does this exceed maximum instructions while the lddw case below does not? Add is one instruction while lddw is two.

[LucasSte]

Ah, Lddw accounts for a single instruction, not two. Let me reformulate my question. Why a does single instruction raise the ExceededMaxInstructions when we only request one unit for execution?

[Lichtso]

lddw only costs one CU. And this tests add, not lddw here.

[LucasSte]

lddw only costs one CU.

I wrote exactly that in my previous comment:

Ah, Lddw accounts for a single instruction, not two

There is no need to repeat what I said. You've probably missed my comment. Now, my question is:

Why a does single instruction raise the ExceededMaxInstructions when we only request one unit for execution?

So in the add test, we only have a single add instruction and request one CU, but we raise an ExceededMaxInstructions error. Why is this the case?

And this tests add, not lddw here.

I didn't understand how this answers my question.

[ripatel-fd]

So in the add test, we only have a single add instruction and request one CU, but we raise an ExceededMaxInstructions error. Why is this the case?

I suppose logically, the ExceededMaxInstructions and ExecutionOverrun faults happen after the same time here, and it's not clear what to return.
After add is executed, the virtual CPU is in a state where it expects another instruction. But at this point CU==0 and PC is OOB, so the next instruction would fault for either of these reasons.

I suppose you could specify that instruction fetch occurs before CU checks, but that's just going to inhibit future possible optimizations.