[rv_dm] dmcontrol.haltreq should always read as 0

[bcstrongx]

desc

The RISC-V spec (v0.13.2) defines the dmcontrol.haltreq bit with access "W" (Section 3.12.2), which is defined as "Write-only. When read this field returns 0" (Table 1.2). However, the current implementation allows this bit to be read as 1 after a write sets it to 1.

[rswarbrick]

I'm looking at the RISC-V external debug support spec, version 0.13.2 (e.g. here).

I think you're right about this behaviour. Looking at the code, the behaviour comes from dm_csrs.sv, where a read of DMControl is handled like this:

        dm::DMControl:    resp_queue_inp.data = dmcontrol_q;

(picking up all the bits of dmcontrol_q, including the halt request).

This code actually comes from the pulp project. I suspect we won't fix it locally (because it doesn't feel very important) and will take a fix by getting it fixed upstream and then vendoring in the change. Would you be happy to file a report there?

[bcstrongx]

Sure, I can file an issue there. Feel free to close this issue if there's no need for it.

[bcstrongx]

Filed, see pulp-platform/riscv-dbg#161

[andreaskurth]

discussed in triage meeting: future release because we don't expect SW to depend on this