Resume "should" use resumereq and halt "should" use haltreq about riscv-debug-spec

[mwachs5]

If it's not clear, we need to clarify it!

Right now the description is here: https://github.com/riscv/riscv-debug-spec/blob/0.13/debug_module.tex#L101

and here:
https://github.com/riscv/riscv-debug-spec/blob/0.13/xml/dm_registers.xml#L89

The intent is that once a hart is halted, it doesn't matter what you write to haltreq (1 or 0). But, if the hart runs again (because you did a resume, or reset) then what is contained in haltreq matters, so you should write it to 0 before doing resumereq (it does say that is undefined), or 1 or 0 during reset depending on what you want to happen when hart comes out of reset.

[poweihuang17]

Thanks for your quick reply. I only looked at the table before. After I see the description in 3.4, it's quite clear to me now.
However, I have a question about resumeack. If a resumeack has no effect on a hart which is running, why could it cause a hart which halts in the future to immediately resume?
https://github.com/riscv/riscv-debug-spec/blob/0.13/debug_module.tex#L129

The spec mention that the debugger should not depend on that behavior, but that sentence(L129) is kind of self-contradictory. Do you feel the same thing?

[mwachs5]

Yes, I think the statement that "resumereq has no effect on hart which is running" is probably the wrong part. I think it should say that debuggers should not do this, because its effect on harts which are running or will halt in the future is undefined. What do you think?

Also, there was another question on the mailing list about this section. I realized the part that is in a comment might be read as referring to what debugger should do on both halt and resume, when it really only means what it should do on resume:
https://github.com/riscv/riscv-debug-spec/blob/0.13/debug_module.tex#L121. We should either clarify or remove this comment.

[mwachs5]

I have attempted to make this more clear in the latest version of the spec.