[QUESTION] How to use interruptible_sleep_on() without races ?

[Jean Tourrilhes <jt@bougret.hpl.hp.com>]

	Hi,

	I've got a problem on using interruptible_sleep_on(). I hope
you will help me fix that ;-)

	I want to wait for a task to finish :
----------------------------------
	if(my_condition == TRUE)
		interruptible_sleep_on(&my_wait_queue);
----------------------------------

	Then, at some point, a timer/BH/soft-irq will do :
-------------------------------
	my_condition == FALSE;
	wake_up_interruptible(&my_wait_queue);
-------------------------------

	It seems straightforward, but it doesn't work. There is a race
condition between the test of the condition and the call to
sleep_on().
	I looked at it in every possible way, and I don't see how it
is possible to use safely interruptible_sleep_on(). And I wonder :
what's the point of having a function in the kernel if you can't use
it safely ?
	As a matter of fact, the TCP code doesn't use
interruptible_sleep_on() but use some complex code around schedule()
(and there must be a simpler and cleaner solution for such a simple
problem).

	Any comments ?

	Jean