On 26/06/2015 18:09, Frederic Konrad wrote:
>>>
>>> +void async_run_safe_work_on_cpu(CPUState *cpu, void (*func)(void
>>> *data),
>>> + void *data)
>>> +{
>> Do you need a mutex to protect this data structure? I would use one
>> even if not strictly necessary, to avoid introducing new BQL-protected
>> structures.
>
> For the moment it's called by tb_invalidate and tb_flush_safe the second
> lacks a
> tb_lock/unlock which should be added. I don't need an other mutex expect
> if this is
> used elsewhere?
In any case, the locking policy should be documented.
At which point you realize that protecting a CPU's
queued_safe_work_{first,next} fields with the tb_lock is a bit weird. :)
I would add a mutex inside CPUState, and then later we could also use
it for regular run_on_cpu/async_run_on_cpu.
Paolo
