Re: [PATCH] 2.5: ewrk3 cli/sti removal by VDA

[Adam Kropelin <akropel1@rochester.rr.com>]

On Fri, Oct 25, 2002 at 06:45:04PM -0400, Jeff Garzik wrote:
> Adam Kropelin wrote:
> >
> >			/* Wait a little while */
> >			spin_unlock_irqrestore(&lp->hw_lock, flags);
> >-			ret = delay;
> >-			__wait_event_interruptible_timeout(wait, 0, ret);
> >+			set_current_state(TASK_INTERRUPTIBLE);
> >+			ret = schedule_timeout(HZ>>2);
> >
> 
> close -- if schedule_timeout() returns greater than zero, that number is 
> the remaining jiffies that schedule_timeout _should_ have slept, but did 
> not.  Ideally you need to call it in a loop, that decrements a variable 
> based on schedule_timeout return code.

My assumption was that the only case in which schedule_timeout() would
return without completing the sleep is if the process was delivered a
signal. So I break the loop immediately in that case. I presume from
your explanation that schedule_timeout() may return for some other reason
(out of curiousity, what?)...and in that case I need to check for a
pending signal, exit if there is one, otherwise schedule_timeout() for
the remaining time. Am I getting warmer?

--Adam