Re: poll() in 2.6 and beyond

[Dave Dillow <dave@thedillows.org>]

On Tue, 2004-03-02 at 16:59, Richard B. Johnson wrote:
> On Tue, 2 Mar 2004, Roland Dreier wrote:

> > I don't know why I continue this, but.... can you point out the line
> > in the kernel 2.4 source for __pollwait() where it sleeps?

> The code in fs/select.c about line 101, adds the current caller
> to the wait-queue. This wait-queue is the mechanism by which the
> current caller sleeps, i.e., gives the CPU up to somebody else.

Actually, no, it does not work that way. The wait queue is the mechanism
by which a process can be *woken up* . The call to add_wait_queue() does
not put the process to sleep.

The process is put to sleep in do_poll(), using schedule_timeout() with
the current state as TASK_INTERRUPTIBLE. This is on line 453.

Your driver will eventually have to wake up sleepers on the queue.
do_poll() will also wake up when the timeout expires, or a signal is
sent to the poll()'ing process.

These semantics have not changed since 2.4. The implementation has
changed a bit, but the driver interface has not changed. Read the
documentation the Roland pointed you to.
-- 
Dave Dillow <dave@thedillows.org>