Re: [PATCH v2 04/17] clocksource: Add Owl timer

[Daniel Lezcano <daniel.lezcano@linaro.org>]

On Sun, Feb 26, 2017 at 03:40:49PM +0100, Andreas Färber wrote:
> Am 25.02.2017 um 22:59 schrieb Daniel Lezcano:
> > On Sat, Feb 25, 2017 at 12:25:32AM +0100, Andreas Färber wrote:
> >> Am 24.02.2017 um 23:29 schrieb Daniel Lezcano:
> >>> On Fri, Feb 24, 2017 at 04:40:42AM +0100, Andreas Färber wrote:
> >>>> +static struct clock_event_device owl_clockevent = {
> >>>> +	.name			= "owl_tick",
> >>>> +	.rating			= 200,
> >>>> +	.features		= CLOCK_EVT_FEAT_ONESHOT,
> >>>
> >>> Did you consider adding CLOCK_EVT_FEAT_DYNIRQ ?
> >>
> >> No, it was not present downstream. Got a good example?
> > 
> > https://lwn.net/Articles/541000/
> 
> Looking at your current Nomadik code, it seems I can literally should
> just add this flag (done), without needing to implement any new hooks.
> 
> On a related topic, how do we determine the cpumask? Downstream and some
> in-tree drivers use cpumask_of(0), others use cpu_possible_mask.

If you specify the CLOCK_EVT_FEAT_DYNIRQ, the cpumask is not important as it
will be changed dynamically.

Otherwise, cpumask_of(0) is often the default because it concentrates the
wakeup on a single cpu, allowing the other cpus to go to deep idle state and if
there are two clusters, it allows to have a cluster idle state. That results on
a better energy saving.

The usage of cpu_possible_mask will randomly wakeup any cpu.
 
> >> Do you spot anything functionally wrong in this driver? Despite adding
> >> this new driver, I am only getting the following additional earlycon output:
> >>
> >> [    0.000029] sched_clock: 32 bits at 24MHz, resolution 41ns, wraps
> >> every 89478484971ns
> >> [    0.007888] clocksource: timer: mask: 0xffffffff max_cycles:
> >> 0xffffffff, max_idle_ns: 79635851949 ns
> >> [    0.017748] Console: colour dummy device 80x30
> >> [    0.022243] Calibrating delay loop...
> >> [    0.030895] random: fast init done
> >> [    0.231021] random: crng init done
> >>
> >> For S900 I'm using the generic timer instead.
> > 
> > I don't get the issue, can you elaborate ?
> 
> Found it myself: I forgot to clear the interrupt pending bit in the
> interrupt handler routine.
> 
> +       writel(OWL_Tx_CTL_PD, owl_timer_base + OWL_T1_CTL);
> 
> Now it goes past this point, initializes the real serial driver and
> boots up to not finding the rootfs:
> 
> [    0.000032] sched_clock: 32 bits at 24MHz, resolution 41ns, wraps
> every 89478484971ns
> [    0.007898] clocksource: timer: mask: 0xffffffff max_cycles:
> 0xffffffff, max_idle_ns: 79635851949 ns
> [    0.017886] Console: colour dummy device 80x30
> [    0.022386] Calibrating delay loop... 405.50 BogoMIPS (lpj=2027520)

May be you should also consider using register_current_timer_delay() instead of
jiffies based delay loops.

> [    0.083523] pid_max: default: 32768 minimum: 301

-- 

 <http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog