Hi,

quoted

+
+#define MTK_RTC_DEV		KBUILD_MODNAME

You probably shouldn't do that and have a static string for the driver
name. I probably doesn't matter much though because DT is used to probe
the driver.

Will change it at next patch.

quoted

+/* we map HW YEAR 0 to 2000 because 2000 is the leap year */
+#define MT2712_MIN_YEAR		2000
+#define MT2712_BASE_YEAR	1900
+#define MT2712_MIN_YEAR_OFFSET	(MT2712_MIN_YEAR - MT2712_BASE_YEAR)
+#define MT2712_MAX_YEAR_OFFSET	(MT2712_MIN_YEAR_OFFSET + 127)
+

All those defines are unecessary, see below.

Will change it at next patch.

quoted

+struct mt2712_rtc {
+	struct device		*dev;

Looking at the code closely, it seems this is only used for debug and
error messages. Maybe you could use rtc_dev->dev instead.

Will change it at next patch.

quoted

+	mutex_lock(&rtc->rtc_dev->ops_lock);
+
+	irqsta = mt2712_readl(rtc, MT2712_IRQ_STA);

Do you have to lock that read? Is the register cleared on read?

Yes, this register is read clear register.

quoted

+	do {
+		__mt2712_rtc_read_time(rtc, tm, &sec);
+	} while (sec < tm->tm_sec);	/* SEC has carried */

Shouldn't that be while (tm->tm_sec < sec)?

In __mt2712_rtc_read_time function, we read tm->tm_sec before read sec.
Sometimes we can meet situation like "tm->tm_sec == 59" and "sec == 0".
It means that TC_SEC has carried and we need to reload the tm struct. I
suppose it was correct that using "while (sec < tm->tm_sec)"

quoted

+
+	/* HW register use 7 bits to store year data, minus
+	 * MT2712_MIN_YEAR_OFFSET brfore write year data to register, and plus
+	 * MT2712_MIN_YEAR_OFFSET back after read year from register
+	 */
+	tm->tm_year += MT2712_MIN_YEAR_OFFSET;

Simply add 100 in __mt2712_rtc_read_time

Will change it at next patch.

quoted

+
+	/* HW register start mon from one, but tm_mon start from zero. */
+	tm->tm_mon--;
+

You can also do that in __mt2712_rtc_read_time.

Will change it at next patch.

quoted

+	if (rtc_valid_tm(tm)) {

This check is unnecessary, the validity is always checked by the core.

Will remove this at next patch.

quoted

+	if (tm->tm_year > MT2712_MAX_YEAR_OFFSET) {
+		dev_dbg(rtc->dev, "Set year %d out of range. (%d - %d)\n",
+			1900 + tm->tm_year, 1900 + MT2712_MIN_YEAR_OFFSET,
+			1900 + MT2712_MAX_YEAR_OFFSET);
+		return -EINVAL;
+	}

This check is unnecessary, see below.

Will change it at next patch.

quoted

+
+	tm->tm_year -= MT2712_MIN_YEAR_OFFSET;
+	tm->tm_mon++;

You should probably avoid modifying tm, move the substraction and
addition in the mt2712_writel calls.

Will change it at next patch.

quoted

+	if (tm->tm_year > MT2712_MAX_YEAR_OFFSET) {
+		dev_dbg(rtc->dev, "Set year %d out of range. (%d - %d)\n",
+			1900 + tm->tm_year, 1900 + MT2712_MIN_YEAR_OFFSET,
+			1900 + MT2712_MAX_YEAR_OFFSET);
+		return -EINVAL;
+	}
+

Unnecessary check.

Will change it at next patch.

quoted

+	p1 = mt2712_readl(rtc, MT2712_POWERKEY1);
+	p2 = mt2712_readl(rtc, MT2712_POWERKEY2);
+	if (p1 != MT2712_POWERKEY1_KEY || p2 != MT2712_POWERKEY2_KEY)
+		dev_dbg(rtc->dev, "powerkey not set (lost power)\n");
+

This info is valuable, you should check that when reading the time and
return -EINVAL if power was lost.

Will change it at next patch.

quoted

+	/* RTC need POWERKEY1/2 match, then goto normal work mode */
+	mt2712_writel(rtc, MT2712_POWERKEY1, MT2712_POWERKEY1_KEY);
+	mt2712_writel(rtc, MT2712_POWERKEY2, MT2712_POWERKEY2_KEY);

This should be written when setting the time after power was lost.

I suppose we can move this into mt2712_rtc_read_time function's "if
(p1 != MT2712_POWERKEY1_KEY || p2 != MT2712_POWERKEY2_KEY)" condition
which will be added at next patch. We need additional flag to mark this
condition or another if condition in mt2712_rtc_set_time fucntion if we
put these code in mt2712_rtc_set_time function.

quoted

+static const struct rtc_class_ops mt2712_rtc_ops = {
+	.read_time	= mt2712_rtc_read_time,
+	.set_time	= mt2712_rtc_set_time,
+	.read_alarm	= mt2712_rtc_read_alarm,
+	.set_alarm	= mt2712_rtc_set_alarm,

For proper operations, you should also provide the .alarm_irq_enable
callback.

Will change it at next patch.

quoted

+	rtc->rtc_dev->ops = &mt2712_rtc_ops;

If you set the range properly here using rtc_dev->range_min and
rtc_dev->range_max, then the core will be able to do range checking and
will also take care of the year offset/windowing calculations instead of
having to hardcode that in the driver.

Will change it at next patch.

Best Regards,
Ran