Scott Johnson

The second statement must take precedent because you can't know if the L bit is set or not until you do the PMP lookup. Keep in mind that an implementation...

> This statement assumes that the "pmp checks" referred to in the first statement are only the permission bits. But if we look at the "checks" to be the combination...

> Two corrections: My step 1 was implicitly assuming the priority was being considered, but I should have stated that. And thanks for finding the text about decomposition, which answers...

> Is it legal behavior for the implementation to trap or not trap in this case? I believe either would be legal behavior. Trap, or not trap. Edit: since you...

+1 on this. Changing a file's permissions when overwriting is not expected behavior.

To solve this robustly, I am thinking of a data structure that holds info about the desired access: * virtual address * effective mode (including 2 bits of privilege and...

`minstret` and `mcycle` have been separated since cd668d2f067d802879a7842bc3bed50fa61b2ead and I can't see where the main loop in `execute.cc` uses the CSR objects to count.

You haven't given us enough information to diagnose the problem. What version of Spike are you using? (Git hash?) What is your host system? (e.g. Ubuntu version) How did you...

It's not clear what the expected behavior vs actual behavior is. My guess: Expected behavior: illegal instruction trap Actual behavior: no illegal instruction trap Is that right?

There is also the debug ROM at 0x0 (see `DEBUG_START` in `riscv/decode.h`) and I suspect that's where the conflict lies. When I run your ELF on a modern Spike, I...