How reliable is the reference model of cosim?

[ha0lyu]

I installed the co-simulation system, followed this tutorial to install cosim and tested it on ibex. One of the test cases shows unexpected behavior:

FAILURE: Co-simulation mismatch at time                 1302
Register write data mismatch to x5 DUT: 2 expected: 1
[1302] %Error: ibex_simple_system_cosim_checker.sv:68: Assertion failed in TOP.ibex_simple_system.u_ibex_simple_system_cosim_checker_bind: Co-simulation mismatch seen
Received stop request, shutting down simulation

I would like to ask, how reliable is the reference model of cosim? I am unsure whether this issue is necessarily a problem with ibex because the reference model version is ibex-cosim-v0.5.

[rswarbrick]

Thanks the question. This looks a lot like it points at a bug (probably in our cosimulation environment). Can you give us step by step instructions about how to reproduce it at our end? (What command, what seed etc.)

[ha0lyu]

Thanks for your reply.

1. Clone ibex, version: 8f4c75c5e
2. Follow the Quick Build and Run Instructions, to build and install the co-simulator
3. build simulator in ibex dir: fusesoc --cores-root=. run --target=sim --setup --build lowrisc:ibex:ibex_simple_system_cosim $(util/ibex_config.py opentitan fusesoc_opts)
4. run cosim: build/lowrisc_ibex_ibex_simple_system_cosim_0/sim-verilator/Vibex_simple_system -t . --meminit=ram,./examples/sw/simple_system/hello_test.elf -c 3000.

Here is the test code:

  asm volatile("li t0, 0x00FF208");
  asm volatile("jalr x0, t0, 0");

Add the code to examples/sw/simple_system/hello_test.c and make, then run ibex.

In fact, this test code will jump to execute at pc=0x000FF208. However, the link.ld file defines the RAM range to start from 0x00100000, and this pc is outside the RAM range. Therefore, the reference model, Spike, throws an Instruction Access Fault trap, while Ibex retrieves the value and decodes it, leading to an Illegal Instruction trap.