RE: condvar wakeups

["Perez-Gonzalez, Inaky" <inaky perez-gonzalez intel com>]

> From: Ingo Molnar [mailto:mingo elte hu]
> 
> here's the latest futex-requeue kernel patch, against 2.5.69. I fixed the
>
> ...
> +			if (++ret <= num) {
> +				wake_up_all_sync(&this->waiters);
> +				if (this->filp)
> +					send_sigio(&this->filp->f_owner,
this->fd, POLL_IN);
> +			} else {
> +				list_add_tail(i, head2);
> +				__attach_vcache(&this->vcache, uaddr2,
current->mm, futex_vcache_callback);
> +				this->offset = offset2;
> +				this->page = page2;
> +			}
> +		}
> +	}
> +
> +	unlock_futex_mm();
> +
> +	unpin_page(page1);
> +	unpin_page(page2);

Call me pedantic, please, but if we are attaching so many waiters 
to a new page, shall not we unpin q->page before requeuing, then
pin that new page once per waiter and then at the tail of futex_wait() 
unpin q->page?

Unless I am missing anything, once we exit from requeue(), page2 is
not pinned by anyone, so it could be perfectly be taken out to swap
space for a walk.

Iñaky Pérez-González -- Not speaking for Intel -- all opinions are my own
(and my fault)