Re: [Qemu-devel] [PATCH] Avoid segfault in cpu_dump_state

[Alexander Graf]

On 23.05.2012, at 17:43, Fabien Chouteau wrote:

> On 05/16/2012 03:39 PM, Fabien Chouteau wrote:
>> On 05/16/2012 10:29 AM, Fabien Chouteau wrote:
>>> On 05/16/2012 05:50 AM, Andreas Färber wrote:
>>>> Am 15.05.2012 18:08, schrieb Fabien Chouteau:
>>>>> On 05/15/2012 03:31 PM, Andreas Färber wrote:
>>>>>> Am 15.05.2012 11:39, schrieb Fabien Chouteau:
>>>>>>> Do not call cpu_dump_state if logfile is NULL.
>>>>>> 
>>>>>> And where is log_cpu_state() being called from? Its caller is passing
>>>>>> NULL already then.
>>>>>> 
>>>>> 
>>>>> No, logfile is a global variable. log_cpu_state() takes only CPUState
>>>>> and flags parameters.
>>>> 
>>>> Ah, I see now that f is a different f here, logfile becomes
>>>> log_cpu_state()'s f. Unfortunate naming.
>>>> 
>>>> Your fix looks OK then but I would recommend turning it into a static
>>>> inline function to avoid the line breaks.
>>>> 
>>> 
>>> In this case I can rewrite all the macros in qemu-log.h to static inline.
>>> 
>> 
>> This is more complex than expected...
>> 
>> 1 - GCC rejects inlined variadic functions
>> 
>> 2 - Moving from macro to inline implies use of types defined in cpu.h
>> (target_ulong, CPUArchState...), which I cannot include because
>> qemu-log.h is used in tools (i.e.  without cpu.h).
>> 
>> Conclusion: unless someone volunteer for a massive restructuring of
>> qemu-log we have to keep the marcro for log_cpu_state.
>> 
> 
> So, are we good with the second patch?

No reply, so I assume that's a yes. Applied it to ppc-next :). Thanks a lot!


Alex