On Thursday 03 August 2006 16:55, Evgeniy Polyakov wrote:
> On Thu, Aug 03, 2006 at 04:40:34PM +0200, Eric Dumazet ([EMAIL PROTECTED])
wrote:
> > > + mutex_lock(&u->ctl_mutex);
> > > + while (num < max_nr && ((k = kqueue_dequeue_ready(u)) != NULL)) {
> > > + if (copy_to_user(buf + num*sizeof(struct ukevent),
> > > + &k->event, sizeof(struct ukevent))) {
> > > + cerr = -EINVAL;
> > > + break;
> > > + }
> >
> > It seems quite wrong to hold ctl_mutex while doing a copy_to_user() (of
> > possibly a large amount of data) : A thread can sleep on a page fault and
> > other threads cannot make progress.
>
> I would not call that wrong - system prevents some threads from removing
> kevents which are counted to be transfered to the userspace, i.e. when
> dequeuing was awakened and it had seen some events it is possible, that
> when it will dequeue them part will be removed by other thread, so I
> prevent this.
Hum, "wrong" was maybe not the good word.... but kqueue_dequeue_ready() uses a
spinlock (ready_lock) to protect ready_list. One particular struct kevent is
given to one thread, one at a time.
If you look at fs/eventpoll.c, you can see how carefull is ep_send_events() so
that multiple threads can in the same time transfer different items to user
memory.
In a model where several threads are servicing events collected by a single
point (epoll, or kevent), this is important to not block all threads because
of a single thread waiting a swapin (trigered by copy_to_user() )
Eric
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
