Draft: refactor the instruction stepper to better support waiting instructions like WFI, WRS.{NTO, STO}

[pmundkur]

This tries to address the issues raised in #398 and implements some ideas discussed in #412.

The first commit refactors the stepper and instruction retirement, and addresses #412.
The second adds a Wait state to the stepper and exposes it to the C emulator.
The third adds a configuration option for the max number of wait states before the Wait (for WFI) expires. This should provide a better foundation for #398.

This is an invasive and backwards-incompatible change, and so mainly submitted for discussion.

I am unsure of the effects on RVFI. Is it still being used?

[github-actions]

Test Results

392 tests  ±0   392 ✅ ±0   1m 22s ⏱️ +2s
  1 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 282e485. ± Comparison against base commit c650166.

♻️ This comment has been updated with latest results.

[jordancarlin]

I am unsure of the effects on RVFI. Is it still being used?

It is still being used. There are plans to transition it to a runtime option instead of compile time, but we definitely don't want to break compatibility with it.

[pmundkur]

What is the recommended way of testing RVFI?

I tried using CTSRD-CHERI/TestRig with utils/scripts/runTestRIG.py -a sail -b toooba -r rv64g. It was able to run on both master and this PR, but errors were reported against both.

[jordancarlin]

What is the recommended way of testing RVFI?

That's a good question and another thing that should be added to CI. Right now we only test building the RVFI model in CI.

[pmundkur]

I confirmed with TestRig that there is no RVFI regression in this PR vs master. The command used was:
utils/scripts/runTestRIG.py -a manual -b manual --implementation-A-port 8080 --implementation-B-port 8081
where the riscv_rvfi_RV64 from the PR and the master were pre-launched with -r <port>

[pmundkur]

@Alasdair I tried out your suggestion to reduce the size of this PR by defining in riscv_inst_retire.sail:

let RETIRE_SUCCESS : Retired(Retire_Failure) = RETIRE_SUCCESS()

But this results in:

Warning: Identifier RETIRE_SUCCESS found in pattern is also a union constructor at riscv_inst_retire.sail:33.4-18:
33 |let RETIRE_SUCCESS : Retired(Retire_Failure) = RETIRE_SUCCESS()
   |    ^------------^
   | 
Suggestion: Maybe you meant to match against RETIRE_SUCCESS() instead?

Type error:
riscv_inst_retire.sail:33.4-18:
33 |let RETIRE_SUCCESS : Retired(Retire_Failure) = RETIRE_SUCCESS()
   |    ^------------^
   | Local variable RETIRE_SUCCESS is already bound as a function name

i.e. it results in both a warning and a type error.

[Alasdair]

You would need to give the constructor a different name.