Branching Speculation Contract Sample Clauses

Branching Speculation Contract. Control flow is the biggest challenge for a complete defi- nition of speculative behavior. Instruction Fetch is the earliest stage of the pipeline and must operate with minimal knowl- edge of the instructions that are ostensibly directing it. Branch prediction is therefore traditionally the most speculative aspect of microarchitecture. In order to define useful constraints for speculative control flow, we strategically limit which instructions will be executed rather than merely fetched. Side channels from speculative execution are useful when they can exfiltrate an illegal value; the production of an illegal value generally happens in the Execute stage of the pipeline, e.g., when a load is actually performed. This allows the front-end of the pipeline to spec- ulate freely until decoding the instruction stream. Allowing wild instruction fetch assumes that it is not possible to leak the contents of memory through the front-end of the pipeline. This should at least require constant-time decoding. After Instruction Decode, the microarchitecture can intelligently enforce architectural speculation contracts with knowledge of the instructions that would be executed. The Branching Speculation Contract (see definition) is de- signed to allow reasoning about the possibilities of speculative execution within a compartment, and also to provide guaran- tees between compartments. BSC item 1 only allows execution to follow the correct target of a direct jump. In this case the target is known in Decode and mispredictions can be flushed before execution. This allows basic blocks to span direct jumps. For example: 1 labelA: 2 jal x0, labelB