Re: [rfc] sk_write_space() for atm

[chas williams <chas@xxxxxxxxxxxxxxxx>]

In message <20030624.174042.27809957.davem@xxxxxxxxxx>,"David S. Miller" writes:
>This looks ok to me, but I am not well versed in this
>area.  For example, if you give a spurious wakeup via
>poll() what can happen?

as far as i can tell this version of poll is fairly close
to datagram_poll so therefore it must be correct :)
i would say the previous version was racy since it needed
to check vcc->reply repeatedly, which could change.  now,
sk_err is checked once and is error state of the socket.   

the only reason i dont use datagram_poll (besides
atm implementing a different writeable condition)
are atm sockets that are waiting (in connecting)
would block when written.  this is similar to a tcp
socket w/o syn having been sent.