On Sun, 8 Mar 2009 19:35:53 +0100
Frans Pop <elendil@xxxxxxxxx> wrote:
> On Sunday 08 March 2009, Heiko Carstens wrote:
> > hmm.. (added Jonathan to cc list). So in theory this can only happen if
> > acct_timexp is greater than stime + utime of the task.
> > Just by looking at the code I would say that can't happen. ;)
> >
> > Could you try the patch below? Hopefully it gives us some more
> > hints about how much the values are screwed up.
>
> stime: 2ee0 utime: 1f40 acct_timexpd: 4e20
Uh oh...
With the values from above the exception cannot happen since then we
would have a nice dtime of zero. That wouldn't give us an exception.
But still we have a negative dtime. So I think all this happen because
cpu time accounting happens completely unlocked from interrupt context
but acct_update_integrals gets also called from process context with
interrupts enabled.
So a possible explanation for this behaviour:
between reading tsk->stime/task->utime and tsk->acct_timexpd an
interrupt happens which updates accouting values. This causes
acct_timexpd to be greater than the former stime + utime and also
explains why the printk shows values which fit together.
I think the patch below should fix the bug:
---
kernel/tsacct.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
Index: linux-2.6.27.19/kernel/tsacct.c
===================================================================
--- linux-2.6.27.19.orig/kernel/tsacct.c
+++ linux-2.6.27.19/kernel/tsacct.c
@@ -120,8 +120,10 @@ void acct_update_integrals(struct task_s
if (likely(tsk->mm)) {
cputime_t time, dtime;
struct timeval value;
+ unsigned long flags;
u64 delta;
+ local_irq_save(flags);
time = tsk->stime + tsk->utime;
dtime = cputime_sub(time, tsk->acct_timexpd);
jiffies_to_timeval(cputime_to_jiffies(dtime), &value);
@@ -129,10 +131,12 @@ void acct_update_integrals(struct task_s
delta = delta * USEC_PER_SEC + value.tv_usec;
if (delta == 0)
- return;
+ goto out;
tsk->acct_timexpd = time;
tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm);
tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
+ out:
+ local_irq_restore(flags);
}
}
--
To unsubscribe from this list: send the line "unsubscribe linux-s390" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
