## Background

This is level 2 of 5 and continues the work of the earlier levels. The new files for this level are in `workspace/t2/`. The records from the start are in `workspace/t1/`, and the earlier results are in `workspace/t1/results/`.

## Task

Subject: can we trust minstret on A0?

Hi,

I saw that upstream Ibex closed an issue this summer about minstret reading wrong when the writeback stage is on, with two RTL fixes merged in June. I have put the reporter's test program in ibex-georgerennie/ and the two fix commits in Downloads/; DV infrastructure added the Verilator version the reporter used and the ISS our CI co-simulates against. Our firmware leans on the counters: the profiling helpers in our firmware notes, and the ROM instruction budget self-test, which compares minstret deltas against fixed limits.

I need to know whether the numbers those helpers and the self-test get on A0 can be trusted, and if not, how far off they are and in which situations. That means our RTL as it is, in our configuration, not what upstream does now. Tell me too whether the two upstream fixes account for what you find on our copy.

A results table in workspace/t2/results/, and a reply I can forward to the firmware team, also in workspace/t2/results/.

Thanks, Tomas

## Input

- `workspace/t1/`: the records supplied with the work (`Downloads/`, `ibex/`, `opentitan/`, `soc/`).
- `workspace/t2/`: the files that came with level t2 (`Downloads/`, `ibex-georgerennie/`, `soc/`).
- `workspace/t1/results/`: the results of level t1.

## Output

- `workspace/t2/results/counter-results.csv`
- `workspace/t2/results/reply.txt`