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

["Andreas Färber" <afaerber@suse.de>]

Am 24.02.2017 um 23:29 schrieb Daniel Lezcano:
> On Fri, Feb 24, 2017 at 04:40:42AM +0100, Andreas Färber wrote:
>> Implement clocksource and clockevents for Actions Semi S500.
>>
>> Based on LeMaker linux-actions tree.
>>
>> Signed-off-by: Andreas Färber <afaerber@suse.de>
> 
> As this is a new driver, please give some technical information about the
> driver itself and a pointer to documentation if it is publicly available.

What technical information would that be? Which of the timers we use as
clock source vs. clock events?

The only public documentation beyond the source tree mentioned is here:

http://www.lemaker.org/product-guitar-download-29.html (section 3.4)

>> +#define OWL_Tx_CTL_INTEN	(1 << 1)
>> +#define OWL_Tx_CTL_EN		(1 << 2)
> 
> s/(1 << 1)/BIT(1)/
> s/(1 << 2)/BIT(2)/

OK

>> +
>> +static void __iomem *owl_timer_base;
>> +
>> +static u64 notrace owl_timer_sched_read(void)
>> +{
>> +	return (u64)readl(owl_timer_base + OWL_T0_VAL);
>> +}
>> +
>> +static int owl_timer_set_state_shutdown(struct clock_event_device *evt)
>> +{
>> +	writel(0, owl_timer_base + OWL_T1_CTL);
>> +
>> +	return 0;
>> +}
>> +
>> +static int owl_timer_set_state_oneshot(struct clock_event_device *evt)
>> +{
>> +	writel(0, owl_timer_base + OWL_T1_CTL);
>> +	writel(0, owl_timer_base + OWL_T1_VAL);
>> +	writel(0, owl_timer_base + OWL_T1_CMP);
>> +
>> +	return 0;
>> +}
>> +
>> +static int owl_timer_tick_resume(struct clock_event_device *evt)
>> +{
>> +	return 0;
>> +}
>> +
>> +static int owl_timer_set_next_event(unsigned long evt,
>> +				    struct clock_event_device *ev)
>> +{
>> +	writel(0, owl_timer_base + OWL_T1_CTL);
>> +
>> +	writel(0, owl_timer_base + OWL_T1_VAL);
>> +	writel(evt, owl_timer_base + OWL_T1_CMP);
>> +
>> +	writel(OWL_Tx_CTL_EN | OWL_Tx_CTL_INTEN, owl_timer_base + OWL_T1_CTL);
>> +
>> +	return 0;
>> +}
>> +
>> +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?

>> +	.set_state_shutdown	= owl_timer_set_state_shutdown,
>> +	.set_state_oneshot	= owl_timer_set_state_oneshot,
>> +	.tick_resume		= owl_timer_tick_resume,
>> +	.set_next_event		= owl_timer_set_next_event,
>> +};
>> +
>> +static irqreturn_t owl_timer_interrupt(int irq, void *dev_id)
>> +{
>> +	struct clock_event_device *evt = (struct clock_event_device *)dev_id;
>> +
>> +	evt->event_handler(evt);
>> +
>> +	return IRQ_HANDLED;
>> +}
>> +
>> +static struct irqaction owl_timer_irq = {
>> +	.name = "owl-timer",
>> +	.flags = IRQF_TIMER,
>> +	.handler = owl_timer_interrupt,
>> +	.dev_id = &owl_clockevent,
>> +};
>> +
>> +static int __init owl_timer_init(struct device_node *node)
>> +{
>> +	const unsigned long rate = 24000000;
> 
> Use DT, either use clock-frequency or a clock ref.

Are clk drivers really available at this point? clock-frequency sounds
more promising.

>> +	int irq1, ret;
>> +
>> +	owl_timer_base = of_io_request_and_map(node, 0, "owl-timer");
>> +	if (IS_ERR(owl_timer_base)) {
>> +		pr_err("Can't map timer registers");
>> +		return -ENXIO;
> 
> Why not PTR_ERR(owl_timer_base) ?

Only one in-tree driver (sun5i) matches such an expression. Will change.

>> +	}
>> +
>> +	irq1 = irq_of_parse_and_map(node, 1);
>> +	if (irq1 <= 0) {
>> +		pr_err("Can't parse timer1 IRQ");
>> +		return -EINVAL;
>> +	}
>> +
>> +	writel(0, owl_timer_base + OWL_T0_CTL);
>> +	writel(0, owl_timer_base + OWL_T0_VAL);
>> +	writel(0, owl_timer_base + OWL_T0_CMP);
>> +	writel(OWL_Tx_CTL_EN, owl_timer_base + OWL_T0_CTL);
> 
> Please factor out these calls into a function.

Do you have something in particular in mind? Maybe ..._reset() for the
first three?

>> +
>> +	sched_clock_register(owl_timer_sched_read, 32, rate);
>> +	clocksource_mmio_init(owl_timer_base + OWL_T0_VAL, node->name,
>> +			      rate, 200, 32, clocksource_mmio_readl_up);
>> +
>> +	writel(0, owl_timer_base + OWL_T1_CTL);
>> +	writel(0, owl_timer_base + OWL_T1_VAL);
>> +	writel(0, owl_timer_base + OWL_T1_CMP);
>> +
>> +	ret = setup_irq(irq1, &owl_timer_irq);
>> +	if (ret) {
>> +		pr_warn("failed to setup irq %d\n", irq1);
>> +		return ret;
>> +	}
> 
> s/setup_irq/request_irq/

Care to explain? setup_irq has 32 hits vs. 28 for request_irq.

>> +
>> +	owl_clockevent.cpumask = cpumask_of(0);
>> +	owl_clockevent.irq = irq1;
>> +
>> +	clockevents_config_and_register(&owl_clockevent, rate,
>> +					0xf, 0xffffffff);
>> +
>> +	return 0;
>> +}
>> +CLOCKSOURCE_OF_DECLARE(owl, "actions,owl-timer", owl_timer_init);

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.

Thanks for your review,

Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)