Re: [RFC] perf: need to expose sched_clock to correlate user samples with kernel samples

[Thomas Gleixner <tglx@linutronix.de>]

On Tue, 19 Feb 2013, John Stultz wrote:
> On 02/19/2013 01:50 PM, Thomas Gleixner wrote:
> > 2) Doing #1 will allow to observe the described time going backwards
> >     scenario in kernel as well.
> > 
> >     The reason why we did not get complaints about that scenario at all
> >     (yet) is that the window and the probability to hit it are small
> >     enough. Nevertheless it's a real issue for virtualized systems.
> > 
> >     Now you came up with the great idea, that the timekeeping core is
> >     able to calculate what the approximate safe value is for the
> >     clocksource readout to be in a state where wreckage relative to the
> >     last update of the clocksource is not observable, not matter how
> >     long the scheduled out delay is and in which direction the NTP
> >     update is going.
> 
> So the other bit of caution here, is I realize my idea of "valid cycle ranges"
> has the potential for deadlock.
> 
> While it should be fine for use with vdso, we have to be careful if we use
> this in-kernel, because if we're in the update path, the valid interval check
> could trigger the ktime_get() in hrtimer_interrupt() to spin forever. So we
> need to be sure we don't use this method anywhere in the code paths that
> trigger the update_wall_time() code.

Hmm, right.
 
> So some additional thinking may be necessary here. Though it may be as simple
> as making sure we don't loop on the cpu that does the timekeeping update.

Either that or make sure to use ktime_get_nocheck() in those code pathes.

Thanks,

	tglx