olofk
commented
2 months ago I haven't found time to look at this issue, but I did just fix an issue that made the regression tests fail. Probably unrelated, but could be worth trying with the latest git version.
Also, during my debugging sessions, I have develeoped some rudimentary tools that can make life slightly easier.
On the trace branch, I have added a WIP tracing module that shows register values and what opcode that was just executed. This makes things slightly easier to follow when doing waveform debugging.
You can also switch on the --trace_pc switch for simulations. This will generate a trace.bin that contains the PC trace. Together with this questionable python hack you can do more analysis of what the CPU is doing. By default it will generate a trace with annotations of all jumps if you run it with python trays.py /path/to/elf/file /path/to/trace.bin
Another thing I typically do is to take the assembler listing of the program you want to run and check for illegal instructions or attempted accesses to unimplemented CSR registers. To save size, SERV has zero protection from these things and will happily try to execute instructions that it doesn't have the slightest clue about how to actually run. Looking quickly at the file you sent me, there is a read from mhartid. That might be totally fine, but worth taking an extra look at. I also see a wfi instruction + a couple of zext.b (which a quick googling tells me is the same as andi??) that could be wort investigating.
m42uko
Hi,
I am currently trying to evaluate the performance of doing some crypto on SERV, but I'm running into some issues with Zephyr on SERV.
I wrote the attached example (main.c) that just initializes the RSA library according to their example, and then prints a string in a loop (no actual en-/decryption happens yet).
Yet, when I run the example (both on real HW and in simulation through fusesoc+verilator), the entire system hangs indefinitely (real HW) or resets (sim). If you remove the mbedtls init calls, the loop runs fine.
I have been trying to troubleshoot this a bit myself already, but I was hoping you might have some more experience analyzing these issues.
I attached the following files: rsa-test-zephyr-serv.zip
Let me know if you have any ideas on how to approach this / if you find out anything. Thank you for your help!
PS: I have not yet tried running the example program on any other RISC-V target. That's still on my TODO list.