Re: [Qemu-devel] Confusion regarding temporaries with branch conditional

[Richard Henderson]

On 11/30/2016 10:12 AM, Alex Bennée wrote:
> 
> Richard Henderson <address@hidden> writes:
> 
>> On 11/30/2016 08:55 AM, Alex Bennée wrote:
>>>
>>> Nikunj A Dadhania <address@hidden> writes:
>>>
>>>> Hi,
>>>>
>>>> I was writing one instruction and hit following issue:
>>>>
>>>> [snip]/qemu/tcg/tcg.c:2039: tcg fatal error
>>>> qemu-ppc64le: [snip]/qemu/translate-all.c:175: tb_lock: Assertion 
>>>> `!have_tb_lock' failed.
>>>> Segmentation fault (core dumped)
>>>
>>> This is confusing because something is trying to take the tb_lock while
>>> you are in code generation. tb_lock is held for code generation to
>>> ensure serialisation of generation.
>>
>> Yes, I've seen this myself.  I never got around to reporting the "problem"
>> properly.  It's a confusing side effect of a SIGSEGV arriving during tcg code
>> generation.  The signal handler longjmps back with unexpected locks
>> held.
> 
> So this is a SEGV which belongs to the translation code rather than the
> guest?

Yes.

> There are places in the cpu loop where we exit that should reset the
> locks on a restart - see tb_lock_reset() so I'm not quite sure what has
> happened here.

It should be easy enough to force a segv from within the compile step to
reproduce this.


r~