Re: Runtime accounting bug?

[Andrew Morton <akpm@linux-foundation.org>]

On Sun, 6 Jul 2008 14:00:22 -0400 (EDT) George Glover <hyperborean@comcast.net> wrote:

> Hello,
> 
> I run two copies (dual processor system) of "mprime" from the GIMPS 
> project. After a while running (weeks?) the cumulative runtime reported by 
> top increments faster than real time, then after a while (unknown how long) 
> the value increments normally again.  Then, maybe something overflows - but 
> runtime accounting stops entirely even though the process is in the run state.
> 
> (mprime is a cpu-bound low priority process like seti@home and friends.)
> 
> I presently have a stuck process and a one that should soon start to 
> increment faster than possible.
> 
> I have verified that the "stuck" process is indeed running since it continues
> to generate output.
> 
> Here is the "stuck" process:
> 
> cat /proc/4126/stat; sleep 5;  cat /proc/4126/stat
> 4126 (mprime) R 2984 4126 2984 34819 4126 4202496 16530 0 4 0 2124505930 661087 0 0 39 19 1 0 8442861 21483520 3733 4294967295 134512640 138881564 3220348480 3220345732 135248565 0 0 0 16386 0 0 0 17 1 0 0 0 0 0
                                                                ^^^^^^^^^^
> 4126 (mprime) R 2984 4126 2984 34819 4126 4202496 16530 0 4 0 2124505930 661087 0 0 39 19 1 0 8442861 21483520 3733 4294967295 134512640 138881564 3220348480 3220345736 135241038 0 0 0 16386 0 0 0 17 1 0 0 0 0 0
                                                                ^^^^^^^^^^

yes?  The 14th column, do_task_stat()'s cputime_to_clock_t(utime)?

> 
> 
> Here is the other process started more recently:
> 
> cat /proc/18312/stat; sleep 5;  cat /proc/18312/stat
> 18312 (mprime) R 2969 18312 2969 34818 18312 4202496 7657 0 1 0 140549087 660363 0 0 39 19 1 0 346388363 35483648 7152 4294967295 134512640 138881564 3221166480 3221163732 135246526 0 0 0 16386 0 0 0 17 1 0 0 0 0 0
                                                                  ^^^^^^^^^
> 18312 (mprime) R 2969 18312 2969 34818 18312 4202496 7657 0 1 0 140549510 660364 0 0 39 19 1 0 346388363 35483648 7152 4294967295 134512640 138881564 3221166480 3221163740 135280398 0 0 0 16386 0 0 0 17 1 0 0 0 0 0
                                                                  ^^^^^^^^^

Here it changed.

> 
> 
> Top output sorted by cpu time:
> 
>    PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+    TIME COMMAND
>   4126 zed       39  19 20980  14m  968 R    0  0.4 354194:30  5903h mprime
> 18312 zed       39  19 34652  27m  952 R   84  0.8  23544:49 392,24 mprime
> 
> 
> uptime:
> 12:50:47 up 60 days, 18:58, 25 users,  load average: 2.20, 2.21, 2.26
> 
> cat /proc/version
> Linux version 2.6.25.1 (root@(none)) (gcc version 4.1.2) #3 SMP PREEMPT Tue
> May 6 01:53:17 EDT 2008
> 
> The machine is a dual processor 1.2Ghz Athlon MP system.  It's generally 
> problem free with maybe 1 bit error a year reported from the ecc ram.
> 
> As the problem takes so long to repeat I do not know how to approach it. 
> It has also been present for the past few kernels, since 2.6.23 if I 
> recall.
> 
> Anyone have any thoughts? It seems more cosmetic than critical.
> 

My suspicion wold be that some of the arithmetic in here:

static cputime_t task_utime(struct task_struct *p)
{
	clock_t utime = cputime_to_clock_t(p->utime),
		total = utime + cputime_to_clock_t(p->stime);
	u64 temp;

	/*
	 * Use CFS's precise accounting:
	 */
	temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);

	if (total) {
		temp *= utime;
		do_div(temp, total);
	}
	utime = (clock_t)temp;

	p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime));
	return p->prev_utime;
}

is suffering overflows at very high values of p->utime.

It's a 32bit machine, yes?

It would useful to know what your values of TICK_NSEC, NSEC_PER_SEC and
USER_HZ are.

Peter, you recently fixed some of that time conversion arithmetic?  But
I think those fixes made it into 2.6.25?