AMD-SB-7005 “Return Address Security Bulletin” outlines this new speculative side channel attack affecting recent EPYC and Ryzen processors.
AMD has received an external report titled ‘INCEPTION’, describing a new speculative side channel attack. AMD believes ‘Inception’ is only potentially exploitable locally, such as via downloaded malware, and recommends customers employ security best practices, including running up-to-date software and malware detection tools. AMD is not aware of any exploit of ‘Inception’ outside the research environment, at this time.
It was done to do the both side of if/else while waiting for the check to finish, then it jumps to the correct branch result execution point to keep going, whatever that was “wrong” is wasted and should be flushed. Don’t know if this fits the modern definition but that’s why they do this type of thing.
Actual design or implementation is more complex than I described as there are a couple ways to tackle this branching delay issue.
Does it execute both, or does it execute the branch that is more likely to be valid? Branch prediction seems like it’d be way more performant than executing both branches until the result of the branch condition is available. If you think about it, what you’re proposing will cause the CPU to always execute instructions that are not meant to be executed when confronted with a branch whereas branch prediction will only execute these “useless” instructions in the unlikely scenario where the prediction is incorrect.
It will prefetch the instructions and put into the pipeline the branch it thinks is mostly likely. It may do ahead-of-time speculative execution on certain instructions but not always. If it missed the correct branch it will flush the pipeline and start the pipeline over again from the correct branch. Afaik it doesn’t execute or prefetch both branches. The other guy is saying it does but that doesn’t really gel with my own recollection or the Wikipedia article he cited. You can see some further discussion that suggests only one branch gets prefetched here here and here. Reasons cited for only predicting one branch are: 1) Two pipelines with all the associated circuitry to look ahead, decode, and speculatively execute is incredibly expensive in terms of both processing requirements and die real estate. 2) Caching both would thrash your caches with new data constantly. 3) Modern branch prediction is already so accurate, there’s really no need for two pipelines anyways.