The decoder is unable to support the Long Address 32-bit ISO Long packets

[MoonLight-SteinsGate]

When I utilize the coresight-decoder to decode the trace data produced by ETMv4 on Juno R2, I find that the decoder can't identify the Long Address 32-bit ISO Long packets with the header 0b10011010.

[MoonLight-SteinsGate]

In addition, the decoder is unable to update the three address registers when decoding the address-related packets.

[retrage]

Thanks for trying out. We tested the tracer and the decoder with ARM64 mainboards listed in the AFL++ CoreSight mode README:
https://github.com/AFLplusplus/AFLplusplus/blob/stable/coresight_mode/README.md
They are not tested under 32-bit address environments yet.
We are always open to any PRs. Please submit a PR to support your desired environment.

[MoonLight-SteinsGate]

In 64-bit address environments, there are some other address-packets such as Exact Match Address, Long Address 64-bit IS1 Long. However, the decoder is unable to recognize these packets. Moreover, ETMv4 utilizes three registers to store the three latest addresses. When the new address-packet is generated, it need to update the registers. However, it seems that this decoder does not support it.

[retrage]

Yes, there are other packets in the ETM specification for 64-bit environments, but CoreSight in our primary experimental environment (Marvell ThunderX2) did not output such packets. This is probably because the CoreSight hardware implementation has not caught up. So we did not implement IS1 Long or anything else that we did not need. If you want a complete decoder, use https://github.com/Linaro/OpenCSD.