On 07/09/2016 08:43, Richard Henderson wrote:
> On 09/06/2016 10:43 PM, Paolo Bonzini wrote:
>>
>>
>> On 07/09/2016 06:03, Richard Henderson wrote:
>>>
>>>> + if (mr->global_locking) {
>>>> + qemu_mutex_lock_iothread();
>>>> + locked = true;
>>>> + }
>>>> memory_region_dispatch_read(mr, physaddr, &val, 1 << SHIFT,
>>>> iotlbentry->attrs);
>>>> + if (locked) {
>>>> + qemu_mutex_unlock_iothread();
>>>> + }
>>>
>>> I'm not keen on this pattern.
>>>
>>> (1) Why not use recursive locks?
>>
>> Probably I'm biased by looking at this code too long, but... how would
>> they help?
>
> You wouldn't check to see if the lock is already taken.
But it's only using qemu_mutex_iothread_locked() at unlock time, when we
exit from longjmp. Here it's looking at a MemoryRegion to see if the
MMIO callbacks need the lock or not (since most of them do).
Paolo
