[NSWI004] Adding print prevents exception 5

[Vojtech Horky]

Hello.

Dne 05. 01. 20 v 21:42 Petr Tůma napsal(a):
> Hi,
> 
> what you describe usually happens when your data structures are corrupt - depending on exact location of things, you may or may not hit a wrong address, and when you update your code and locations change, the exact place where the exception occurs can also move.
> 
> I would probably try to look at the addresses in thread and as, and at the address that causes the exception, perhaps that would help. We can also take a look, but please commit the code and provide us with exact commit hash and exact steps to reproduce the problem.

Printing thread->as would show that it is a NULL pointer. Looking at the 
code, I would suspect that thread->as initialization does not work as 
expected for the very first thread.

The weird behavior is probably a combination of multiple factors. As a 
best guess without diving too much into the code, it is quite possible 
that the NULL pointer dereference is actually translated to a physical 
address zero (due to ERL == 1 handling). Therefore you are overwriting 
your exception handling routines (their assembler parts).


Another hint is the panic print from your exception code:

Kernel panic: Exception...5, status: 0x80005364, epc: 0

* EPC 0 means you would be returning to execute code at address where 
code is never placed
* STATUS looks too much like an address rather than status code

Both these things suggest corruption of the exception handling routines.


BTW, have you tried running GDB with a breakpoint inside thread_create?


Hope this helps,
- VH


> 
> By the way, writing code such as ++p->q->r is quite dangerous, few people know C operator precedence and associativity rules well enough to read this with any certainty. Writing ++(p->q->r) is more readable.
> 
> Petr
> 
> 
> On 04/01/2020 21:15, Tomáš Drozdík wrote:
>> Hi,
>>
>> I've encountered an interesting problem that I cannot really reason about.
>>
>> As I was dealing with counting references to an address space my code
>> suddenly started throwing Exception 5. Via debug prints I was able to
>> track down when this exception is thrown. What came as a surprise is
>> that if a place a print right before this instruction everything works
>> properly (i.e. tests which should pass pass). Moving this print
>> anywhere else ruins it and exception is thrown.
>>
>> This is a little snippet from thread_create function:
>> ```
>> printk("REMOVE THIS PRINT AND EXCEPTION 5 IS THROWN\n");
>> ++thread->as->reference_counter;
>> ```
>> I've looked up that Exception 5 is an store to illegal address. What
>> does it have to do with print is beyond me, obviously adding print
>> does shift code in the memory and so it may get aligned or something.
>> But alignment of code is compilers job. Therefore illegal address
>> should be in memory, possibly in as_t structure but that contains just
>> 4 integers.
>> I don't know how much code I can paste here and even if I could I'm
>> not even remotely aware of what can cause such behavior.
>>
>> Link to repo: https://gitlab.mff.cuni.cz/teaching/nswi004/winter-2019-20/team-i_cant_c/tree/as5
>> Commit hash: 80761a068cd7f342d42e7786eaf5611419ac0487
>>
>> Any help would be really appreciated.
>> Thanks in advance.
>>
>> Sincerly
>> Tomáš Drozdík
>> _______________________________________________
>> NSWI004 mailing list
>> 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
>