Simulation gets stuck with SIM=verilator 4.106 about cocotbext-axi HOT 10 OPEN

@andreaskuster I was also facing this issue, where after I upgraded several libraries the simulation for my previously working project got stuck. I'm using Verilator 4.106, and was using a previous release of cocotb and cocotbext-axi. It turns out the issue was with cocotb-bus, where earlier I was using version 0.1.1 and after the upgrade, it was 0.2.1.

Reverting cocotb-bus to version 0.1.1worked for me. And this works with even the newest cocotb and cocotb-ext versions.

pip3 install cocotb-bus==0.1.1

When a bus version > 0.1.1 is used the reset never gets asserted. I don't know why reset is affected. But hopefully, this solution helps.

from cocotbext-axi.

Hi @andreaskuster Only to clarify, I am simply a recent user of cocotbext-axi.

I can't right now, but I will try to reproduce it and let you know. Let me say that I ran verilator --version in my Ubuntu 20.04 and it reported Verilator 4.028 2020-02-06 rev v4.026-92-g890cecc1 (are you sure that you are trying 1.106? seems extremely old).

from cocotbext-axi.

Thanks for your reply. And yes you were right, i meant version 4.106 (fixed it in the description), and the compatibility issue with newer versions i mentioned is according to the cocotb github issues and the comment here: https://docs.cocotb.org/en/stable/simulator_support.html

Furthermore, i tested different (older) versions of cocotbext-axi, with the same behaviour, so this problem seems to be present for a while.

from cocotbext-axi.

Ok, I didn't know about the cocotb limitation using Verilator. I generally use cocotb with GHDL or iverilog.

from cocotbext-axi.

Sure, ghdl and iverilog are fine, as long as you work with vhdl or verilog. However, I am currently extending a project written in SystemVerilog, which is why I would like to get verilator up to speed.

Well, it is not really a limitation by design, but rather the devil that comes with the fact that verilator is extending the VPI infrastructure. This should eventually become compatible again.

However, I am not sure if verilator is really to blame here, since the latest (v1.6.1) cocotb basic examples seem to work with verilator v4.106. However, it is difficult for me to debug the issue, since this co-routine based code seems to be stuck in an infinity loop, but I do not understand the cause yet. @alexforencich Please let me know if you have any insights or clues that I can help with.

from cocotbext-axi.

I'm not sure what the problem is here. But I'm also not going to do anything with verilator until verilator/verilator#2778 is resolved so that cocotb will work correctly with current versions of verilator.

from cocotbext-axi.

Yeah sure I can understand that. Thanks for your clarification. Let's keep this open and come back as soon as verilator has been fixed.

BTW: cocotbext-axi + questasim works well for systemverilog.

from cocotbext-axi.

I mean, it's also possible that I introduced some sort of bug recently that only affects verilator. So it's probably still worth getting to the bottom of what's going on here. Previously, I think I have tested this on verilator 4.106, and it worked fine. But there have been some changes since then, although I can't think of any change in particular that might have affected verilator specifically.

from cocotbext-axi.

I think that the problem either exists for a longer time, or it is already broken in verilator 4.106, since I tried different, older version combinations of cocotbext-axi and cocotb, but with the very same result of getting stuck.

from cocotbext-axi.

Thanks @SabraO, this indeed did the trick!

I opened a PR #37 to fix this.

from cocotbext-axi.