linux-kernel - Re: [PATCH v4 4/7] rtc: spacemit: support the SpacemiT P1 RTC

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <a99ad08a-99b2-4e39-a2d1-28302d312735@riscstar.com>
Date: Wed, 25 Jun 2025 17:10:54 -0500
From: Alex Elder <elder@...cstar.com>
To: Alexandre Belloni <alexandre.belloni@...tlin.com>
Cc: lee@...nel.org, lgirdwood@...il.com, broonie@...nel.org, robh@...nel.org,
 krzk+dt@...nel.org, conor+dt@...nel.org, mat.jonczyk@...pl, dlan@...too.org,
 paul.walmsley@...ive.com, palmer@...belt.com, aou@...s.berkeley.edu,
 alex@...ti.fr, troymitchell988@...il.com, linux-rtc@...r.kernel.org,
 devicetree@...r.kernel.org, linux-riscv@...ts.infradead.org,
 spacemit@...ts.linux.dev, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v4 4/7] rtc: spacemit: support the SpacemiT P1 RTC

On 6/25/25 5:01 PM, Alexandre Belloni wrote:
> On 25/06/2025 11:41:15-0500, Alex Elder wrote:
>> +/*
>> + * The P1 hardware documentation states that the register values are
>> + * latched to ensure a consistent time snapshot within the registers,
>> + * but these are in fact unstable due to a bug in the hardware design.
>> + * So we loop until we get two identical readings.
>> + */
>> +static int p1_rtc_read_time(struct device *dev, struct rtc_time *t)
>> +{
>> +	struct p1_rtc *p1 = dev_get_drvdata(dev);
>> +	struct regmap *regmap = p1->regmap;
>> +	u32 count = RTC_READ_TRIES;
>> +	u8 times[2][6];
>> +	size_t size;
>> +	u32 i = 0;
>> +	u8 *time;
>> +	int ret;
>> +
>> +	size = sizeof(times[0]);
>> +	ret = regmap_bulk_read(regmap, RTC_TIME, times[i], size);
>> +	if (ret)
>> +		return ret;
>> +
>> +	do {
>> +		i = 1 - i;
>> +		ret = regmap_bulk_read(regmap, RTC_TIME, times[i], size);
>> +		if (ret)
>> +			return ret;
>> +	} while (memcmp(times[i], times[1 - i], size) && --count);
> 
> Simply checking the seconds is enough, unless you expect regmap_bulk_read to ake
> more than a minute so you don't need a two dimension array.

I hadn't thought it through, but you're right.  I'll still
do bulk reads but will save only the seconds register and
compare it to the previous.

> 
>> +
>> +	if (!count)
>> +		return -EIO;
>> +
>> +	time = &times[0][0];
>> +
>> +	t->tm_sec = time[0] & GENMASK(5, 0);
>> +	t->tm_min = time[1] & GENMASK(5, 0);
>> +	t->tm_hour = time[2] & GENMASK(4, 0);
>> +	t->tm_mday = (time[3] & GENMASK(4, 0)) + 1;
>> +	t->tm_mon = time[4] & GENMASK(3, 0);
>> +	t->tm_year = (time[5] & GENMASK(5, 0)) + 100;
>> +
>> +	return 0;
>> +}
>> +
>> +/*
>> + * The P1 hardware documentation states that values in the registers are
>> + * latched so when written they represent a consistent time snapshot.
>> + * Nevertheless, this is not guaranteed by the implementation, so we must
>> + * disable the RTC while updating it.
>> + */
>> +static int p1_rtc_set_time(struct device *dev, struct rtc_time *t)
>> +{
>> +	struct p1_rtc *p1 = dev_get_drvdata(dev);
>> +	struct regmap *regmap = p1->regmap;
>> +	u8 time[6];
>> +	int ret2;
>> +	int ret;
>> +
>> +	time[0] = t->tm_sec;
>> +	time[1] = t->tm_min;
>> +	time[2] = t->tm_hour;
>> +	time[3] = t->tm_mday - 1;
>> +	time[4] = t->tm_mon;
>> +	time[5] = t->tm_year - 100;
>> +
>> +	/* Disable the RTC to update; re-enable again when done */
>> +	ret = regmap_update_bits(regmap, RTC_CTRL, RTC_EN, 0);
>> +	if (ret)
>> +		return ret;
>> +
>> +	ret = regmap_bulk_write(regmap, RTC_TIME, time, sizeof(time));
>> +
> Honnestly, I'd simply go for
> 
> 	if (ret)
> 		return ret;
> 
> Here, you are trying to set the time, I'm not sure it is worth trying to
> reenable the rtc while hoping the previous time has been kept.

OK.  I normally at least try to return to the original state
even when errors occur.

I don't know who updates the RTC, but I suppose if any error
occurs writing a value they should assume it's "not working".

> This also shows that p1_rtc_read_time should check that RTC_EN is set, else it
> knows the time has never been set and is invalid.

That's a good point.  I'll add a check.  And with that it should
be fine to return without re-enabling.

Thanks for your review.  I'll wait until tomorrow to send the
next version.

					-Alex

>> +	ret2 = regmap_update_bits(regmap, RTC_CTRL, RTC_EN, RTC_EN);
>> +
>> +	return ret ? : ret2;		/* Return the first error */
> 
> 
>> +}
>> +
>> +static const struct rtc_class_ops p1_rtc_class_ops = {
>> +	.read_time = p1_rtc_read_time,
>> +	.set_time = p1_rtc_set_time,
>> +};
>> +
>> +static int p1_rtc_probe(struct platform_device *pdev)
>> +{
>> +	struct device *dev = &pdev->dev;
>> +	struct rtc_device *rtc;
>> +	struct p1_rtc *p1;
>> +
>> +	p1 = devm_kzalloc(dev, sizeof(*p1), GFP_KERNEL);
>> +	if (!p1)
>> +		return -ENOMEM;
>> +	dev_set_drvdata(dev, p1);
>> +
>> +	p1->regmap = dev_get_regmap(dev->parent, NULL);
>> +	if (!p1->regmap)
>> +		return dev_err_probe(dev, -ENODEV, "failed to get regmap\n");
>> +
>> +	rtc = devm_rtc_allocate_device(dev);
>> +	if (IS_ERR(rtc))
>> +		return dev_err_probe(dev, PTR_ERR(rtc),
>> +				     "error allocating device\n");
>> +	p1->rtc = rtc;
>> +
>> +	rtc->ops = &p1_rtc_class_ops;
>> +	rtc->range_min = RTC_TIMESTAMP_BEGIN_2000;
>> +	rtc->range_max = RTC_TIMESTAMP_END_2063;
>> +
>> +	clear_bit(RTC_FEATURE_ALARM, rtc->features);
>> +	clear_bit(RTC_FEATURE_UPDATE_INTERRUPT, rtc->features);
>> +
>> +	return devm_rtc_register_device(rtc);
>> +}
>> +
>> +static struct platform_driver p1_rtc_driver = {
>> +	.probe = p1_rtc_probe,
>> +	.driver = {
>> +		.name = MOD_NAME,
>> +	},
>> +};
>> +
>> +module_platform_driver(p1_rtc_driver);
>> +
>> +MODULE_DESCRIPTION("SpacemiT P1 RTC driver");
>> +MODULE_LICENSE("GPL");
>> +MODULE_ALIAS("platform:" MOD_NAME);
>> -- 
>> 2.45.2
>>
>