Re: Jiffies wraparound is not treated in the schedstats

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Mauricio Lin wrote:
> For instance the delta_jiffies variable is simply calculated as:
> 
> delta_jiffies = now - t->sched_info.last_queued;
>
> Do not you think the more logical way should be

No. Learn about the modulo arithmetic properties of the unsigned types.
 
> if (time_after(now, t->sched_info.last_queued))

Have you looked at time_after? It just performs a subtraction, after
casting both operands to long. It does not care which one is bigger than
the other. Basically, it performs the same calculation as the
delta_jiffies above, and then checks the sign bit.

>    delta_jiffies = now - t->sched_info.last_queued;
> else
>    delta_jiffies = (MAX_JIFFIES - t->sched_info.last_queued) + now

I'm looking at a 2.6.17 kernel, and I can't find MAX_JIFFIES, or
JIFFIES_MAX. There is a MAX_JIFFY_OFFSET, which is something else. It's
the biggest delta that you can add to a value X so that time_after(X, X
+ delta) is still true. If you try to schedule something beyond now +
MAX_JIFFY_OFFSET, it will be put back in time.

Also, your calculation is buggy. MAX_JIFFIES would correspond to
ULONG_MAX. What you want is:

  (ULONG_MAX + 1 - last_queued + now)

This expression is now mathematically congruent to 

  -last_queued + now (mod ULONG_MAX + 1)

or

   now - last_queued (mod ULONG_MAX + 1).

The arithmetic of the unsigned long type is /already/ carried out modulo
ULONG_MAX + 1, so the ULONG_MAX + 1 term can be dropped from your
expression, leaving it as

  now - last_queued

> I have included more variables to measure some issues of schedule in
> the kernel (following schedstat idea) and I noticed that jiffies
> wraparound has led to wrong values, since the user space tool when
> collecting the values is producing negative values.

What user space tool?

> Any comments?

The user space tool is perhaps buggy, or you are misinterpreting its
output.
 
> Can I provide a patch for that?

Some day, when you learn how to program.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux