openocd not honouring watchpoint hit time (before/after) in mcontrol6

[RaamyPi]

When using mcontrol6, spike breaks after the watchpoint is hit. But openocd only checks if the watchpoint is hit and not the other bit also which indicates when the said watchpoint was hit. Subsequently, openocd assumes that the watchpoint is always hit before and sets the step bit in dcsr which ultimately results in control being returned after one more instruction.

[en-sc]

... and not the other bit also which indicates when the said watchpoint was hit.

OpenOCD checks the whole mcontrol6.hit field:

riscv-openocd/src/target/riscv/riscv.c

Lines 1630 to 1632 in 7b4ad6f

	case CSR_TDATA1_TYPE_MCONTROL6:
	hit_mask = CSR_MCONTROL6_HIT0 | CSR_MCONTROL6_HIT1;
	break;

riscv-openocd/src/target/riscv/riscv.c

Lines 1649 to 1657 in 7b4ad6f

	if (tdata1 & hit_mask) {
	LOG_TARGET_DEBUG(target, "Trigger %u (unique_id=%" PRIi64 ") has hit bit set.",
	i, r->trigger_unique_id[i]);
	if (riscv_reg_set(target, GDB_REGNO_TDATA1, tdata1 & ~hit_mask) != ERROR_OK)
	return ERROR_FAIL;
	*unique_id = r->trigger_unique_id[i];
	break;
	}

Subsequently, openocd assumes that the watchpoint is always hit before and sets the step bit in dcsr which ultimately results in control being returned after one more instruction.

Yes, OpenOCD always steps from a breakpoint or a watchpoint. There is #1144 aiming to address this issue.

Moreover, I'm not checking hit field to skip the step worth it. The thing is, hit field is optional. Therefore, making sure OpenOCD does step from a trigger when the trigger was hit with a timing before requires a lot of communication with the target and maybe always stepping is simpler and faster.