[NSWI004] Stuck on syscall return

[Vojtech Horky]

Hello.

Dne 16. 02. 20 v 16:27 Patrik Smelik napsal(a):
> Sorry, forgot to mention, the branch test1 is the one i'm currently 
> using. Well im working on as6 but didn't want to push it yet since it 
> doesn't really work.

Please, bottom/in-line post :-).

> 
> So we don't need to jump again, that's great. Wasn't sure if it is 
> needed or not, but it looked like it got better with the jump.

Syscall is a exception and the jump back is handled for all exceptions 
in the same way in the assembler code we provided.


> But without the jump, the syscall is called again for some reason and it 
> tries to print again. And looks like it tries to print something, it 
> doesn't have access to, since it crashes there. But this time it causes 
> a kernel panic loop (the one my colleague mention earlier this week). 
> With this im even more confused.

And which test have you tried to run?

Looking at your code, I do not see where you distinguish between puts 
and putc in libc syscalls...

HTH,
- VH




> 
> - PS
> 
> ne 16. 2. 2020 o 16:03 Vojtech Horky <horky at d3s.mff.cuni.cz 
> <mailto:horky at d3s.mff.cuni.cz>> napísal(a):
> 
>     Hello.
> 
>     Dne 16. 02. 20 v 15:25 Patrik Smelik napsal(a):
>      > Hi,
>      >
>      > Since yesterday, we are stuck on this problem. Userspace process is
> 
>     It would be extremely useful to send commit hash where the bug is
>     reproducible, your master is four weeks old, I do not see
>     implementation
>     of print in branch as6...
> 
> 
>      > created fine, syscall handling also works, e.g. for print it
>     prints out
>      > what it needs to, but when returning from syscall, i.e. jumping
>     back to
>      > userspace, the application suddenly terminates. I fed the
>      > cpu_jump_to_userspace function context->sp and context->epc as
>      > arguments.The problem is that the context->sp is set to 0 for some
>      > reason, which i believe isn't the correct input for the jump. I
>     traced
>      > the change to the memcpy method, so to when the binary image is
>     being
>      > copied, but i'm currently stuck here.
> 
>     Not sure I got it right, but when the syscall is handled, the thread
>     returns to userspace normally as with any other exception. That is, you
>     need to call cpu_jump_to_userspace only once per process.
> 
>     And you only use context->epc to advance to next instruction as
>     otherwise the syscall would be restarted after being handled.
> 
>     Also note that this function would typically receive two constants:
>     address of the top of stack (so something like 0x3f00) and entry point
>     as specified in the linker script (0x4000).
> 
>     As a side note: in one of your commits I see that you panic when the
>     syscall id is invalid. That is definitely a bad behavior because any
>     application would be easily able to terminate your kernel.
> 
>     Hope this helps,
>     - VH
>     _______________________________________________
>     NSWI004 mailing list
>     NSWI004 at d3s.mff.cuni.cz <mailto:NSWI004 at d3s.mff.cuni.cz>
>     https://d3s.mff.cuni.cz/mailman/listinfo/nswi004
> 
> 
> _______________________________________________
> NSWI004 mailing list
> NSWI004 at d3s.mff.cuni.cz
> https://d3s.mff.cuni.cz/mailman/listinfo/nswi004
>