[ofa-general] Re: wait_for_completion_timeout() spurious failure under heavy load?

[Ingo Molnar]

* Jiri Slaby <jirislaby at gmail.com> wrote:

>> 		do {
>> 			timeout = schedule_timeout(timeout);
>> 	
>> 			if (!timeout)
>> 				return timeout;
>> 	
>> 		} while (!x->done);
>> 	
>> 		return timeout;
>> 	}
>>
>> so if the system is very busy and x->done is not set when
>> do_wait_for_common() is entered, it is possible that the first call to
>> schedule_timeout() returns 0 because the task doing wait_for_completion
>
> Sorry, but how can schedule_timeout return 0 before the timeout 
> expiration?

the point would be that due to high load, the completion wakeup happens 
first, but then due to scheduling delays the timeout also occurs 
(later), before the wakeup related to x->done has managed to do its 
task.

I.e. due to scheduling delays we report a spurious "timeout" failure, 
despite the completion occuring before the timeout. The timeout is 
really intended to be related to the delay of the completion event, not 
the delay of scheduling after that event happened.

seems like a well-spotted race to me, i agree it's more robust to ignore 
the timeout if we can make progress on x->done, and return a 1 jiffy 
'timeout remaining' value. Oleg, do you concur?

but i'd do it not like this:

>                       if (!timeout) {
>                               __remove_wait_queue(&x->wait, &wait);
> -                             return timeout;
> +                             if (x->done) {
> +                                     x->done--;
> +                                     return 1;
> +                             } else {
> +                                     return 0;
> +                             }

but like in the commit below. Agreed?

	Ingo

-------------------->
commit bb10ed0994927d433f6dbdf274fdb26cfcf516b7
Author: Roland Dreier <rdreier at cisco.com>
Date:   Thu Jun 19 15:04:07 2008 -0700

    sched: fix wait_for_completion_timeout() spurious failure under heavy load
    
    It seems that the current implementaton of wait_for_completion_timeout()
    has a small problem under very high load for the common pattern:
    
    	if (!wait_for_completion_timeout(&done, timeout))
    		/* handle failure */
    
    because the implementation very roughly does (lots of code deleted to
    show the basic flow):
    
    	static inline long __sched
    	do_wait_for_common(struct completion *x, long timeout, int state)
    	{
    		if (x->done)
    			return timeout;
    
    		do {
    			timeout = schedule_timeout(timeout);
    
    			if (!timeout)
    				return timeout;
    
    		} while (!x->done);
    
    		return timeout;
    	}
    
    so if the system is very busy and x->done is not set when
    do_wait_for_common() is entered, it is possible that the first call to
    schedule_timeout() returns 0 because the task doing wait_for_completion
    doesn't get rescheduled for a long time, even if it is woken up early
    enough.
    
    In this case, wait_for_completion_timeout() returns 0 without even
    checking x->done again, and the code above falls into its failure case
    purely for scheduler reasons, even if the hardware event or whatever was
    being waited for happened early enough.
    
    It would make sense to add an extra test to do_wait_for() in the timeout
    case and return 1 if x->done is actually set.
    
    A quick audit (not exhaustive) of wait_for_completion_timeout() callers
    seems to indicate that no one actually cares about the return value in
    the success case -- they just test for 0 (timed out) versus non-zero
    (wait succeeded).
    
    Signed-off-by: Ingo Molnar <mingo at elte.hu>

diff --git a/kernel/sched.c b/kernel/sched.c
index 4a3cb06..577f160 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -4405,6 +4405,16 @@ do_wait_for_common(struct completion *x, long timeout, int state)
 			spin_unlock_irq(&x->wait.lock);
 			timeout = schedule_timeout(timeout);
 			spin_lock_irq(&x->wait.lock);
+
+			/*
+			 * If the completion has arrived meanwhile
+			 * then return 1 jiffy time left:
+			 */
+			if (x->done && !timeout) {
+				timeout = 1;
+				break;
+			}
+
 			if (!timeout) {
 				__remove_wait_queue(&x->wait, &wait);
 				return timeout;