Re: [Qemu-devel] [PULL 1/2] util/aio-win32: Only select on what we are actually waiting for

[Alistair Francis]

On Mon, Jul 17, 2017 at 7:23 PM, Paolo Bonzini <address@hidden> wrote:
>> +        if (io_read) {
>> +            bitmask |= FD_READ | FD_ACCEPT | FD_CLOSE;
>> +        }
>> +
>> +        if (io_write) {
>> +            bitmask |= FD_WRITE | FD_CONNECT;
>> +        }
>> +
>>          event = event_notifier_get_handle(&ctx->notifier);
>> -        WSAEventSelect(node->pfd.fd, event,
>> -                       FD_READ | FD_ACCEPT | FD_CLOSE |
>> -                       FD_CONNECT | FD_WRITE | FD_OOB);
>> +        WSAEventSelect(node->pfd.fd, event, bitmask);
>>      }
>>
>
> As noticed by Eric, if the same socket is in use via both aio-win32 and
> another GSource, or via multiple AioContexts, this would break because
> there is only one WSAEventSelect mask for each socket handle.
>
> It is probably not going to break anything in the case of aio-win32, but I
> think it is worth at least a comment. And because WSAEventSelect is
> edge-triggered (e.g. an FD_WRITE event doesn't trigger again until the next
> write) it shouldn't be that bad for performance.  If there's time to revert
> this patch, I think it would be preferable.

It looks like this was applied, I'll be back in the office next week
should I try and revert this or just add a comment explaining the
limitation.

Either that or store the current mask and OR then all together.

Thanks,
Alistair

>
> Paolo