Re: [RFC PATCH] timekeeping: introduce timekeeping_is_busy()

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

Hi John,

* John Stultz (john.stultz@linaro.org) wrote:
> On 09/11/2013 08:08 AM, Mathieu Desnoyers wrote:
> > Starting from commit 06c017fdd4dc48451a29ac37fc1db4a3f86b7f40
> > "timekeeping: Hold timekeepering locks in do_adjtimex and hardpps"
> > (3.10 kernels), the xtime write seqlock is held across calls to
> > __do_adjtimex(), which includes a call to notify_cmos_timer(), and hence
> > schedule_delayed_work().
> >
> > This introduces a side-effect for a set of tracepoints, including mainly
> > the workqueue tracepoints: a tracer hooking on those tracepoints and
> > reading current time with ktime_get() will cause hard system LOCKUP such
> > as:
> Oh bummer. I had just worked this issue out the other day:
>     https://lkml.org/lkml/2013/9/9/476
>
> Apparently it was a schroedinbug of sorts. My apologies for time you
> spent chasing this down.

No worries. As soon as I've been able to reproduce it on my test box
(with serial port), the NMI watchdog had a pretty reasonable explanation
for the issue.

> My plan is to pull the notify_cmos_timer call to outside of the
> timekeeper locking (see the patch at the very end of the mail in the
> above link), as well as try to add lockdep support to seqcount/seqlocks
> so we can catch these sorts of issues more easily.

I just tried your patch, and it indeed seems to fix the lockup I've been
experiencing with lttng-modules. Do you plan pushing this fix into
master, and submitting it for inclusion into stable 3.10 and stable 3.11 ?
I'm planning on dealing with with this issue with a blacklist of
kernel versions that will prevent people from building lttng-modules
against broken kernels as soon as the patch makes it into those trees.

I will reply to the rest of your email separately, so this thread can
focus on getting the fix upstream quickly.

Thanks!

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com