lkml.org 
[lkml]   [2013]   [Dec]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 4/9] clocksource/cadence_ttc: Use enable/disable_irq
Hi Thomas,

sorry for the delay, but I couldn't find time any earlier.

On Thu, Nov 28, 2013 at 08:07:10PM +0100, Thomas Gleixner wrote:
> On Thu, 28 Nov 2013, Sören Brinkmann wrote:
> > On Thu, Nov 28, 2013 at 03:18:50PM +0100, Thomas Gleixner wrote:
> > > Now the problem with this device is that it is not a per cpu
> > > device. It's a global device, so this update can conflict with a
> > > parallel access on the other CPU. Now the disable_irq() only prevents
> > > that the other CPU can handle a device interrupt from that timer. But
> > > it does not prevent any parallel access from e.g. the idle code path
> > > which will try to reprogram it.
> >
> > Does that mean interrupts need to be disabled globally? Also, does the
>
> Globally disabling interrupts is not going to work, except you want to
> use stomp_machine(). But that would be overkill.
>
> > cpuidle path depend on interrupts or can it interfere no matter what?
>
> It can interfere no matter what. The broadcast is modified for the cpu
> which loses its per cpu timer due to the idle state via
>
> clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, ...);
>
> > > Soren, is that timer used as the broadcast device ?
> >
> > Yes, this is the only broadcast capable timer on Zynq, AFAIK. Other than
> > the TTC we only have the arm_global_timer and smp_twd timers, which both
> > are per_cpu devices and thus not broadcast capable.
>
> There is a solution to this. We can identify the broadcast device in
> the core and serialize all callers including interrupts on a different
> cpu against the update. So no need for the disable/enable_irq() dance.

IIUC, and please correct me if I'm wrong, with the patch I'd simply call
'clockevents_update_freq() without having to disable IRQs. But I'm not
sure whether periodic mode is covered. I found, that I had to reprogram
the timer interval in my clock notifier callback when the timer
frequency changes. I think 'clockevents_update_freq()' only handles
oneshot mode. For that reason I call 'ttc_set_interval()' in the clock
notifier in case the timer is in periodic mode. For that call we'd still
have possible races. I guess the best solution would be to move that
functionality into 'clockevents_update_freq()'?

Sören


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2013-12-07 02:21    [W:0.064 / U:0.412 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site