Re: [PATCH 1/1] eventfd: implementation of EFD_MASK flag

[Martin Sustrik <sustrik@250bpm.com>]

On 08/02/13 07:36, Andy Lutomirski wrote:

>> On 08/02/13 02:03, Andy Lutomirski wrote:
>>>
>>> There may be some
>>> advantage to adding (later on, if needed) an option to change the
>>> flags set in:
>>>
>>> +               if (waitqueue_active(&ctx->wqh))
>>> +                       wake_up_locked_poll(&ctx->wqh,
>>> +                               (unsigned long)ctx->mask.events);
>>>
>>> (i.e. to allow the second parameter to omit some bits that were
>>> already signaled.)  Allowing write to write a bigger struct in the
>>> future won't break anything.
>>
>>
>> I think I don't follow. Either the second parameter is supposed to be
>> *newly* signaled events, in which case the events that were already signaled
>> in the past should be ommitted, or it is meant to be *all* signaled events,
>> in which case the current implementation is OK.
>
> I defer to the experts here.  But I suspect that if you want to
> perfectly emulate sockets, you may need to vary what you specify.
> (IIRC tcp sockets report an EPOLLIN edge every time data is received
> even if the receive buffer wasn't empty.)

Hm. That sounds like leaking protocol implementation details to the 
user. That's a bad design IMO and should not be encouraged.

Anyway, I have implemented your other suggestions.

Btw, one thing I am not sure about is how to submit improved patches to 
the ML. Should I use the same patch name? Doesn't that cause confusion?

Martin