[PATCH v3 3/5] ARM: bcm4760: Add system timer

[cavokz@gmail.com (Domenico Andreoli)]

On Wed, Aug 14, 2013 at 05:30:38PM -0700, Stephen Boyd wrote:
> On 08/15, Domenico Andreoli wrote:
> > +
> > +static inline void __iomem *to_load(struct bcm4760_timer *timer)
> > +{
> > +	return timer->base + TIMER_LOAD_OFFSET;
> > +}
> > +
> > +static inline void __iomem *to_control(struct bcm4760_timer *timer)
> > +{
> > +	return timer->base + TIMER_CONTROL_OFFSET;
> > +}
> > +
> > +static inline void __iomem *to_intclr(struct bcm4760_timer *timer)
> > +{
> > +	return timer->base + TIMER_INTCLR_OFFSET;
> > +}
> > +
> > +static inline void __iomem *to_ris(struct bcm4760_timer *timer)
> > +{
> > +	return timer->base + TIMER_RIS_OFFSET;
> > +}
> > +
> > +static inline void __iomem *to_mis(struct bcm4760_timer *timer)
> > +{
> > +	return timer->base + TIMER_MIS_OFFSET;
> > +}
> 
> Style Nit: This is new. Usually people either make a
> <my_driver>_{readl,writel}() function that takes the struct and
> an offset or they just add the offset directly in their
> readl/writel calls. Can you do that? Probably save some lines of
> code.

yes, sure.

> 
> > +static irqreturn_t bcm4760_timer_interrupt(int irq, void *dev_id)
> > +{
> > +	struct bcm4760_timer *timer = dev_id;
> > +	void (*event_handler)(struct clock_event_device *);
> > +
> > +	/* check the (masked) interrupt status */
> > +	if (!readl_relaxed(to_mis(timer)))
> > +		return IRQ_NONE;
> > +
> > +	/* clear the timer interrupt */
> > +	writel_relaxed(1, to_intclr(timer));
> > +
> > +	event_handler = ACCESS_ONCE(timer->evt.event_handler);
> > +	if (event_handler)
> > +		event_handler(&timer->evt);
> 
> This is unfortunate. Do you have a pending timer interrupt left
> by the bootloader?

Do you mean that if no interrupts are expected beween the irq request and
the call clockevents_config_and_register(), I can assume event_handler is
always set?

> 
> > +
> > +	return IRQ_HANDLED;
> > +}
> > +
> > +static void __init bcm4760_init_time(struct device_node *node)
> > +{
> > +	void __iomem *base;
> > +	u32 freq = 24000000;
> 
> Why have freq in the DT binding at all then?

To get an HW address and remap it at runtime, I guess. I copied it but
would also prefer to use DT only where really needed.

> 
> > +	int irq;
> > +	struct bcm4760_timer *timer;
> > +
> > +	base = of_iomap(node, 0);
> > +	if (!base)
> > +		panic("Can't remap timer registers");
> > +
> > +	timer = kzalloc(sizeof(*timer), GFP_KERNEL);
> > +	if (!timer)
> > +		panic("Can't allocate timer struct\n");
> > +
> > +	irq = irq_of_parse_and_map(node, 0);
> > +	if (irq <= 0)
> > +		panic("Can't parse timer IRQ");
> > +
> > +	timer->base = base;
> > +	timer->evt.name = node->name;
> > +	timer->evt.rating = 300;
> > +	timer->evt.features = CLOCK_EVT_FEAT_ONESHOT;
> > +	timer->evt.set_mode = bcm4760_timer_set_mode;
> > +	timer->evt.set_next_event = bcm4760_timer_set_next_event;
> > +	timer->evt.cpumask = cpumask_of(0);
> > +	timer->act.name = node->name;
> > +	timer->act.flags = IRQF_TIMER | IRQF_SHARED;
> > +	timer->act.dev_id = timer;
> > +	timer->act.handler = bcm4760_timer_interrupt;
> > +
> > +	if (setup_irq(irq, &timer->act))
> > +		panic("Can't set up timer IRQ\n");
> > +
> > +	clockevents_config_and_register(&timer->evt, freq, 0xf, 0xffffffff);
> 
> If you switch this registration and the setup_irq() call you
> shouldn't need the ACCESS_ONCE() and that check in the irq handler.
> Please switch the order and or clear the interrupt before
> registering the clockevent and remove the checks in the interrupt
> handler.

got it. Thank you.

Domenico