[PATCH 2/4] drivers/rtc/rtc-armada38x: Add a new RTC driver for recent mvebu SoCs

Hi Arnaud,

[...]

quoted

+struct armada38x_rtc {
+	struct rtc_device   *rtc_dev;
+	void __iomem	    *regs;
+	void __iomem	    *regs_soc;
+	spinlock_t	    lock;

out of curiosity, why use a spinlock and not a mutex?

To be able to use it in the interrupt context.

quoted

+	int		    irq;
+};
+
+/*
+ * According to the datasheet, we need to wait for 5us between each
+ * write
+ */
+static void rtc_delayed_write(u32 val, struct armada38x_rtc *rtc, int offset)
+{
+	writel(val, rtc->regs + offset);
+	udelay(5);
+}

In the remaining of the driver, you have various direct calls to writel()
w/o that 5?s protection, to update/modifiy rtc registers. Is that 5?s
trick specific to a given subpart of the registers? In that case, I
guess it would help to update the comment.

This direct call were done because there was not a call to write just after.
As explained in the comment the 5us is needed only _between_ two consecutive
write. I can emphasize it needed.

quoted

+
+static int armada38x_rtc_read_time(struct device *dev, struct rtc_time *tm)
+{
+	struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+	unsigned long time, time_check, flags;
+
+	spin_lock_irqsave(&rtc->lock, flags);
+
+	time = readl(rtc->regs + RTC_TIME);
+	/*
+	 * WA for failing time set attempts. As stated in HW ERRATA if
+	 * more than one second between two time reads is detected
+	 * then read once again.
+	 */
+	time_check = readl(rtc->regs + RTC_TIME);
+	if ((time_check - time) > 1)
+		time_check = readl(rtc->regs + RTC_TIME);

Bear with me; I don't have access to HW ERRATA.

Are you guaranteed to get a valid value at third read if you got a bad
one before, i.e. no need to put a while loop around that workaround?

I don't have enough information to answer you, maybe the Marvell designer can
answer it. I will ask.

Additionally, the workaround seems to be related to time
setting. Looking at time setting code below, it seems the issue is also
created by the fact the writel() call is not under the protection of the
spinlock.

quoted

+
+	spin_unlock_irqrestore(&rtc->lock, flags);
+
+	rtc_time_to_tm(time_check, tm);
+
+	return 0;

Does the block provide anyway to detect the oscillator was not running,
i.e. the value we are reading is not valid?

I didn't see anything related to this, but here again I will ask.

quoted

+}
+
+static int armada38x_rtc_set_time(struct device *dev, struct rtc_time *tm)
+{
+	struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+	int ret = 0;
+	unsigned long time;
+
+	ret = rtc_tm_to_time(tm, &time);

quoted

+
+	if (ret)
+		goto out;
+	/*
+	 * Setting the RTC time not always succeeds. According to the
+	 * errata we need to first write on the status register and
+	 * then wait for 1s before writing to the time register to be
+	 * sure that the data will be taken into account.
+	 */
+	writel(0, rtc->regs + RTC_STATUS);
+	ssleep(1);
+
+	writel(time, rtc->regs + RTC_TIME);

I think writel(time + 1, ...) would correct the one second shift you
introduce using you ssleep() above.

I will just move the rtc_tm_to_time-) function here, before the second write.

Additionally, as discussed above, I do not see why you're not protecting
the write using you spinlock.

Initially the spinlock was around the 2 writel, but you can't have sleep
in a critical section hold by spinlock so I removed it. However it would be
possible to use them around each writel, but I am not sure that it makes sens.

quoted

+
+out:
+	return ret;
+}
+
+static int armada38x_rtc_read_alarm(struct device *dev, struct rtc_wkalrm *alrm)
+{
+	struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+	unsigned long time, flags;
+	u32 val;
+
+	spin_lock_irqsave(&rtc->lock, flags);
+
+	time = readl(rtc->regs + RTC_ALARM1);
+	val = readl(rtc->regs + RTC_IRQ1_CONF) & RTC_IRQ1_AL_EN;
+
+	spin_unlock_irqrestore(&rtc->lock, flags);
+
+	alrm->enabled = val ? 1 : 0;
+	rtc_time_to_tm(time,  &alrm->time);
+
+	return 0;
+}
+

In the two functions below, regarding RTC interrupt activation, do you
need a check that a valid IRQ was passed in the .dts and that
devm_request_irq() went ok in probe(), i.e. that rtc->irq > 0?

You're right is the irq is not valid, I can even removed these 2 functions
from armada38x_rtc_ops.

quoted

+static int armada38x_rtc_set_alarm(struct device *dev, struct rtc_wkalrm *alrm)
+{
+	struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+	unsigned long time, flags;
+	int ret = 0;
+	u32 val;
+
+	ret = rtc_tm_to_time(&alrm->time, &time);
+
+	if (ret)
+		goto out;
+
+	spin_lock_irqsave(&rtc->lock, flags);
+
+	rtc_delayed_write(time, rtc, RTC_ALARM1);
+
+	if (alrm->enabled) {
+			rtc_delayed_write(RTC_IRQ1_AL_EN , rtc, RTC_IRQ1_CONF);
+			val = readl(rtc->regs_soc + SOC_RTC_INTERRUPT);
+		writel(val | SOC_RTC_ALARM1_MASK,
+			rtc->regs_soc + SOC_RTC_INTERRUPT);
+	}
+
+	spin_unlock_irqrestore(&rtc->lock, flags);
+
+out:
+	return ret;
+}
+
+static int armada38x_rtc_alarm_irq_enable(struct device *dev,
+					 unsigned int enabled)
+{
+	struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&rtc->lock, flags);
+
+	if (enabled)
+		writel(RTC_IRQ1_AL_EN, rtc->regs + RTC_IRQ1_CONF);
+	else
+		writel(0, rtc->regs + RTC_IRQ1_CONF);

I find the following more readable, but it is subjective:

writel(enabled ? RTC_IRQ1_AL_EN : 0, rtc->regs + RTC_IRQ1_CONF);

quoted

+
+	spin_unlock_irqrestore(&rtc->lock, flags);
+
+	return 0;
+}
+
+static irqreturn_t armada38x_rtc_alarm_irq(int irq, void *data)
+{
+	struct armada38x_rtc *rtc = data;
+	u32 val;
+	int event = RTC_IRQF | RTC_AF;
+
+	dev_dbg(&rtc->rtc_dev->dev, "%s:irq(%d)\n", __func__, irq);
+	val = readl(rtc->regs_soc + SOC_RTC_INTERRUPT);
+
+	spin_lock(&rtc->lock);
+
+	writel(val & ~SOC_RTC_ALARM1, rtc->regs_soc + SOC_RTC_INTERRUPT);

Why not putting the readl() above under protection of the spinlock to
protect a change that could occur between the readl() and the writel()?

Indeed it would be better.

quoted

+	val = readl(rtc->regs + RTC_IRQ1_CONF);
+	/* disable all the interrupts for alarm 1 */
+	rtc_delayed_write(0, rtc, RTC_IRQ1_CONF);
+	/* Ack the event */
+	writel(RTC_STATUS_ALARM1, rtc->regs + RTC_STATUS);
+
+	spin_unlock(&rtc->lock);
+
+	if (val & RTC_IRQ1_FREQ_EN) {
+		if (val & RTC_IRQ1_FREQ_1HZ)
+			event |= RTC_UF;
+		else
+			event |= RTC_PF;
+	}
+
+	rtc_update_irq(rtc->rtc_dev, 1, event);
+
+	return IRQ_HANDLED;
+}
+
+static const struct rtc_class_ops armada38x_rtc_ops = {
+	.read_time = armada38x_rtc_read_time,
+	.set_time = armada38x_rtc_set_time,
+	.read_alarm = armada38x_rtc_read_alarm,
+	.set_alarm = armada38x_rtc_set_alarm,
+	.alarm_irq_enable = armada38x_rtc_alarm_irq_enable,
+};
+
+static __init int armada38x_rtc_probe(struct platform_device *pdev)
+{
+	struct resource *res;
+	struct armada38x_rtc *rtc;
+	int ret;
+
+	rtc = devm_kzalloc(&pdev->dev, sizeof(struct armada38x_rtc),
+			    GFP_KERNEL);
+	if (!rtc)
+		return -ENOMEM;
+
+	spin_lock_init(&rtc->lock);
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	rtc->regs = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(rtc->regs))
+		return PTR_ERR(rtc->regs);
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	rtc->regs_soc = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(rtc->regs_soc))
+		return PTR_ERR(rtc->regs_soc);
+
+	rtc->irq = platform_get_irq(pdev, 0);
+
+	if (rtc->irq < 0) {
+		dev_err(&pdev->dev, "no irq\n");
+		return rtc->irq;
+	}
+	if (devm_request_irq(&pdev->dev, rtc->irq, armada38x_rtc_alarm_irq,
+				0, pdev->name, rtc) < 0) {
+		dev_warn(&pdev->dev, "Interrupt not available.\n");
+		rtc->irq = -1;
+	}
+	platform_set_drvdata(pdev, rtc);
+
+	device_init_wakeup(&pdev->dev, 1);

if devm_request_irq() returns an error, should device_init_wakeup() be
called? See comment below under armada38x_rtc_suspend().

We should not called it and also set the set_alarm and alarm_irq_enable operation
to NULL.

quoted

+	rtc->rtc_dev = devm_rtc_device_register(&pdev->dev, pdev->name,
+					&armada38x_rtc_ops, THIS_MODULE);
+	if (IS_ERR(rtc->rtc_dev)) {
+		ret = PTR_ERR(rtc->rtc_dev);
+		dev_err(&pdev->dev, "Failed to register RTC device: %d\n", ret);
+		if (ret == 0)
+			ret = -EINVAL;

I may be missing something but I do not see how ret can be 0 above
because the initial check uses IS_ERR() and not IS_ERR_OR_NULL().

Indeed I can remove this. The devm_rtc_device_register won't return NULL.

quoted

+		return ret;
+	}
+	return 0;
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int armada38x_rtc_suspend(struct device *dev)
+{
+	if (device_may_wakeup(dev)) {
+		struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+
+		return enable_irq_wake(rtc->irq);
+	}

AFAICT, rtc->irq maybe -1 in the call above. You need to either call
device_init_wakeup() when devm_request_irq() succeed or check rtc->irq
is valid in the "if" above.

I won't call  device_init_wakeup()  if rtc->irq is -1.

Thanks,

Gregory

quoted

+
+	return 0;
+}
+
+static int armada38x_rtc_resume(struct device *dev)
+{
+	if (device_may_wakeup(dev)) {
+		struct armada38x_rtc *rtc = dev_get_drvdata(dev);
+
+		return disable_irq_wake(rtc->irq);
+	}
+
+	return 0;
+}
+#endif
+
+static SIMPLE_DEV_PM_OPS(armada38x_rtc_pm_ops,
+			 armada38x_rtc_suspend, armada38x_rtc_resume);
+
+#ifdef CONFIG_OF
+static const struct of_device_id armada38x_rtc_of_match_table[] = {
+	{ .compatible = "marvell,armada-380-rtc", },
+	{}
+};
+#endif
+
+static struct platform_driver armada38x_rtc_driver = {
+	.driver		= {
+		.name	= "armada38x-rtc",
+		.pm	= &armada38x_rtc_pm_ops,
+		.of_match_table = of_match_ptr(armada38x_rtc_of_match_table),
+	},
+};
+
+module_platform_driver_probe(armada38x_rtc_driver, armada38x_rtc_probe);
+
+MODULE_DESCRIPTION("Marvell Armada 38x RTC driver");
+MODULE_AUTHOR("Gregory CLEMENT [off-list ref]");
+MODULE_LICENSE("GPL");

-- 
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com