Re: [RFC PATCH v1 15/31] ARC: Process/scheduling/clock/Timers/Delay Management

[Thomas Gleixner <tglx@linutronix.de>]

On Wed, 7 Nov 2012, Vineet Gupta wrote:
> +void cpu_idle(void)
> +{
> +	/* Since we SLEEP in idle loop, TIF_POLLING_NRFLAG can't be set */
> +
> +	/* endless idle loop with no priority at all */
> +	while (1) {
> +		tick_nohz_idle_enter();
> +
> +		while (!need_resched())
> +			arch_idle();
> +
> +		tick_nohz_idle_exit();
> +
> +		preempt_enable_no_resched();
> +		schedule();
> +		preempt_disable();

  		schedule_preempt_disabled() please

> +	}

> diff --git a/arch/arc/kernel/time.c b/arch/arc/kernel/time.c
> 
> +static void arc_periodic_timer_setup(unsigned int limit)
> +{
> +	/* setup start and end markers */
> +	write_aux_reg(ARC_REG_TIMER0_LIMIT, limit);
> +	write_aux_reg(ARC_REG_TIMER0_CNT, 0);	/* start from 0 */
> +
> +	/* IE: Interrupt on count = limit,
> +	 * NH: Count cycles only when CPU running (NOT Halted)
> +	 */
> +	write_aux_reg(ARC_REG_TIMER0_CTRL, TIMER_CTRL_IE | TIMER_CTRL_NH);
> +}
> +
> +/*
> + * Acknowledge the interrupt & enable/disable the interrupt
> + */
> +static void arc_periodic_timer_ack(unsigned int irq_reenable)
> +{
> +	/* 1. Ack the interrupt by writing to CTRL reg.
> +	 *    Any write will cause intr to be ack, however it has to be one of
> +	 *    writable bits (NH: Count when not halted)
> +	 * 2. If required by caller, re-arm timer to Interrupt at the end of
> +	 *    next cycle.
> +	 *
> +	 * Small optimisation:
> +	 * Normal code would have been
> +	 *  if (irq_reenable) CTRL_REG = (IE | NH); else CTRL_REG = NH;
> +	 * However since IE is BIT0 we can fold the branch
> +	 */
> +	write_aux_reg(ARC_REG_TIMER0_CTRL, irq_reenable | TIMER_CTRL_NH);
> +}

....

> +/********** Clock Event Device *********/
> +
> +static int arc_clkevent_set_next_event(unsigned long delta,
> +				    struct clock_event_device *dev)
> +{
> +	arc_periodic_timer_setup(delta);

This is confusing. Is arc_periodic_timer_setup() setting up a periodic
timer or a oneshot timer? It looks you use it for both and the
differentiation happens in arc_periodic_timer_ack(). So I assume the
timer only knows about periodic mode, but you trick it into oneshot
with the ack function, right ? So it's just me being confused about
the function names, but that could do with some explanatory comments.

> +	return 0;
> +}
> +
> +static void arc_clkevent_set_mode(enum clock_event_mode mode,
> +			       struct clock_event_device *dev)
> +{
> +	pr_info("Device [%s] clockevent mode now [%d]\n", dev->name, mode);

Please remove the debug leftover.

> +	switch (mode) {
> +	case CLOCK_EVT_MODE_PERIODIC:
> +		arc_periodic_timer_setup(CONFIG_ARC_PLAT_CLK / HZ);
> +		break;
> +	case CLOCK_EVT_MODE_ONESHOT:
> +		break;
> +	default:
> +		break;
> +	}
> +
> +	return;
> +}
> +
> +static DEFINE_PER_CPU(struct clock_event_device, arc_clockevent_device) = {
> +	.name		= "ARC Timer0",
> +	.features	= CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_PERIODIC,
> +	.mode		= CLOCK_EVT_MODE_UNUSED,
> +	.rating		= 300,
> +	.irq		= TIMER0_IRQ,	/* hardwired, no need for resources */
> +	.set_next_event = arc_clkevent_set_next_event,
> +	.set_mode	= arc_clkevent_set_mode,
> +};
> +
> +irqreturn_t timer_irq_handler(int irq, void *dev_id)

static please

> +static int arc_finished_booting;
> +
> +/*
> + * Scheduler clock - returns current time in nanosec units.
> + * It's return value must NOT wrap around.
> + *
> + * Although the return value is nanosec units based, what's more important
> + * is whats the "source" of this value. The orig jiffies based computation
> + * was only as granular as jiffies itself (10ms on ARC).
> + * We need something that is more granular, so use the same mechanism as
> + * gettimeofday(), which uses ARC Timer T1 wrapped as a clocksource.
> + * Unfortunately the first call to sched_clock( ) is way before that subsys
> + * is initialiased, thus use the jiffies based value in the interim.
> + */
> +unsigned long long sched_clock(void)
> +{
> +	if (!arc_finished_booting) {
> +		return (unsigned long long)(jiffies - INITIAL_JIFFIES)
> +		    * (NSEC_PER_SEC / HZ);
> +	} else {
> +		struct timespec ts;
> +		getrawmonotonic(&ts);

This can live lock. sched_clock() is used by the tracer. So assume you
are function tracing and you trace a function called from within the
timekeeping seqcount write "locked" region. You spin forever in
getrawmonotonic(). Not what you want, right ?

Thanks,

	tglx