On Mon, 27 Mar 2017 09:56:47 +0800
Wanpeng Li <kernel...@gmail.com> wrote:
> Actually after I bisect, the first bad commit is ff9a9b4c4334 ("sched,
> time: Switch VIRT_CPU_ACCOUNTING_GEN to jiffy granularity"). The bug
> can be reproduced readily if CONFIG_CONTEXT_TRACKING_FORCE is true,
> then just stress all the online cpus or just one cpu and leave others
> idle(so it stresses the global timekeeping one), top show 100%
> sys-time. And another way to reproduce it is by nohz_full, and gives
> the stress to the house keeping cpu, the top show 100% sys-time of the
> house keeping cpu, and also the other cpus who have at least two tasks
> running on and in full_nohz mode.
We're not short on reproducers, I have a new one too:
http://people.redhat.com/~lcapitul/real-time/acct-bug.c
This is a single threaded task that reproduces the issue. If you
run it as instructed, you'll get:
- nohz_full CPU: 95% system time 5% idle time
- non-nohz_full CPU: 95% user time 5% idle time (expected behavior)
This reproduces the issue, but not for the reasons I expected. I was
trying to mimic what I was seeing on my trace when tracing the two
task problem. Which is: a task stays 995us in user-space and then
enters the kernel. Time won't be accounted for user-space because
we're not 1 jiffies yet, but if the task stays in the kernel for more
than 5us, then time will be accounted for system time when going
back to user-space.
However, what really seems to be happening is: acct-bug is causing
the tick to be re-activated (why? it shouldn't) and that causes the
issue to appear. This is consistent with my other observations: I
can only reproduce the issue if the nohz_full CPU re-activates the tick.
> Let's consider the cpu which has responsibility for the global
> timekeeping, as the tracing posted above, the vtime_account_user() is
> called before tick_sched_timer() which will update jiffies,
But the vtime_account_user() call and the jiffies update happen
on different CPUs, no? So the ordering shouldn't matter.
> so jiffies
> is stale in vtime_account_user() and the run time in userspace is
> skipped, the vtime_user_enter() is called after jiffies update, so
> both the time in userspace and in kernel are accumulated to sys time.
>
> If the housekeeping cpu is idle when CONFIG_NO_HZ_FULL, everything is
> fine. However, if you give stress to the housekeeping cpu, top will
> show 100% sys-time of both the housekeeping cpu and the other cpus who
> have at least two tasks running on and in full_nohz mode.
The housekeeping CPUs are idle with my reproducers.
> I think it
> is because the stress delays the timer interrupt handling in some
> degree, then the jiffies is not updated timely before other cpus
> access it in vtime_account_user().
>
> I think we can keep syscalls/exceptions context tracking still in
> jiffies based sampling and utilize local_clock() in vtime_delta()
> again for irqs which avoids jiffies stale influence. I can make a
> patch if the idea is acceptable or there is any better proposal. :)
>
> Regards,
> Wanpeng Li
>
