The ExtendedDTraceProbes flag enables method entry/exit probes and allocation probes for generated code. Currently C2 calls these through stubs which is unnecessary and also forces slow path allocation for allocation notification. This makes the generated code run extremely slowly, i.e from 5 to 30 times slower than without the probes. Switching to real leaf calls and allowing fast path allocation reduces the slowdown to around 2 to 3 times which is still high but more manageable. Additionally if we could allow inlining the cost would be much less but inlining may interact badly with the jstack action which doesn't understand inlining. If we upgrade the jstack action we could consider enabling inlining.