[PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[Arun Iyengar via B4 Relay]

From: Arun Iyengar <aiyengar1990@gmail.com>

For an RTC with RTC_FEATURE_ALARM_RES_MINUTE, rtc_set_alarm() rounds the
alarm down to the start of its minute, so that it never fires late. When
the alarm is less than a minute away and in the same minute as the
current time, the rounded time is already in the past.
rtc_timer_enqueue() then gets -ETIME from __rtc_set_alarm() and expires
the timer from the irq work right away, and the hardware alarm is never
armed.

That loses every short wake alarm. With a HYM8563, which has minute
resolution, "rtcwake -m freeze -s 15" sets an alarm that expires in
software before the system suspends, so nothing wakes it: with the
alarm set at 17:55:00 for 17:55:17, /proc/driver/rtc reads back
"alrm_time: 17:55:00, alarm_IRQ: no", and the system stays in
suspend-to-idle until something else wakes it.

Round the alarm up to the next minute when rounding down would put an
alarm that is still in the future into the past. It then fires at most
59 seconds late instead of never. Alarms that are already in the past,
or that round down to a time still in the future, are handled as before.

Fixes: d87f741dddab ("rtc: handle alarms with a minute resolution")
Cc: stable@vger.kernel.org
Signed-off-by: Arun Iyengar <aiyengar1990@gmail.com>
---
Tested on an Orange Pi 5 (RK3588S) with a HYM8563: "rtcwake -m no -s 15"
now reads back the alarm at the next full minute with "alarm_IRQ: yes",
and 47 consecutive "rtcwake -m freeze -s 15" cycles all woke from the
RTC interrupt. Before the fix the alarm was lost whenever the 15 s
stayed inside the current minute, and the system never woke.

This patch was developed with Claude (Anthropic) as a coding assistant:
it wrote the code and the commit message, and I reviewed and tested it
as described above.
---
 drivers/rtc/interface.c | 33 +++++++++++++++++++++++++++------
 1 file changed, 27 insertions(+), 6 deletions(-)

diff --git a/drivers/rtc/interface.c b/drivers/rtc/interface.c
index 96626f8068f9..004819c9a846 100644
--- a/drivers/rtc/interface.c
+++ b/drivers/rtc/interface.c
@@ -535,18 +535,39 @@ int rtc_set_alarm(struct rtc_device *rtc, struct rtc_wkalrm *alarm)
 		rtc_timer_remove(rtc, &rtc->aie_timer);
 
 	alarm_time = rtc_tm_to_ktime(alarm->time);
-	/*
-	 * Round down so we never miss a deadline, checking for past deadline is
-	 * done in __rtc_set_alarm
-	 */
-	if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features))
-		alarm_time = ktime_sub_ns(alarm_time, (u64)alarm->time.tm_sec * NSEC_PER_SEC);
+	if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features) &&
+	    alarm->time.tm_sec) {
+		ktime_t down = ktime_sub_ns(alarm_time,
+					    (u64)alarm->time.tm_sec * NSEC_PER_SEC);
+		struct rtc_time tm;
+		ktime_t now;
+
+		err = __rtc_read_time(rtc, &tm);
+		if (err)
+			goto unlock;
+		now = rtc_tm_to_ktime(tm);
+
+		/*
+		 * Round down so we never miss a deadline. If that puts an alarm
+		 * that is still in the future into the past, the timer would
+		 * expire right away and the hardware alarm would never be
+		 * armed: whoever relies on it, e.g. to wake the system from
+		 * suspend, would wait forever. Round up to the next minute
+		 * instead. An alarm already in the past is left alone, it is
+		 * handled by __rtc_set_alarm.
+		 */
+		if (ktime_after(down, now) || !ktime_after(alarm_time, now))
+			alarm_time = down;
+		else
+			alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);
+	}
 
 	rtc->aie_timer.node.expires = alarm_time;
 	rtc->aie_timer.period = 0;
 	if (alarm->enabled)
 		err = rtc_timer_enqueue(rtc, &rtc->aie_timer);
 
+unlock:
 	mutex_unlock(&rtc->ops_lock);
 
 	return err;

---
base-commit: 7485659ecb81d5fce3a3edf2cae9fcb29c17b1cd
change-id: 20261110-rtc-minute-alarm-1d8af9689d23

Best regards,
--  
Arun Iyengar <aiyengar1990@gmail.com>

Re: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [High] rtc: 32-bit signed long overflow in 60 * NSEC_PER_SEC in
  rtc_set_alarm()
--

commit 4251dabe39af884c0054960d9c538feb4fe7bb5e
Author: Arun Iyengar <aiyengar1990@gmail.com>

rtc: Don't lose a minute-resolution alarm that rounds down into the past

This commit rounds minute-resolution RTC alarms up to the next minute when
rounding down to the start of the minute would move a future alarm into the
past.

> diff --git a/drivers/rtc/interface.c b/drivers/rtc/interface.c
> index 96626f8068f93..004819c9a846f 100644
> --- a/drivers/rtc/interface.c
> +++ b/drivers/rtc/interface.c
> @@ -535,18 +535,39 @@ int rtc_set_alarm(struct rtc_device *rtc, struct rtc_wkalrm *alarm)
>  		rtc_timer_remove(rtc, &rtc->aie_timer);
>  
>  	alarm_time = rtc_tm_to_ktime(alarm->time);
> -	/*
> -	 * Round down so we never miss a deadline, checking for past deadline is
> -	 * done in __rtc_set_alarm
> -	 */
> -	if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features))
> -		alarm_time = ktime_sub_ns(alarm_time, (u64)alarm->time.tm_sec * NSEC_PER_SEC);
> +	if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features) &&
> +	    alarm->time.tm_sec) {
> +		ktime_t down = ktime_sub_ns(alarm_time,
> +					    (u64)alarm->time.tm_sec * NSEC_PER_SEC);
> +		struct rtc_time tm;
> +		ktime_t now;
> +
> +		err = __rtc_read_time(rtc, &tm);
> +		if (err)
> +			goto unlock;
> +		now = rtc_tm_to_ktime(tm);
> +
> +		/*
> +		 * Round down so we never miss a deadline. If that puts an alarm
> +		 * that is still in the future into the past, the timer would
> +		 * expire right away and the hardware alarm would never be
> +		 * armed: whoever relies on it, e.g. to wake the system from
> +		 * suspend, would wait forever. Round up to the next minute
> +		 * instead. An alarm already in the past is left alone, it is
> +		 * handled by __rtc_set_alarm.
> +		 */
> +		if (ktime_after(down, now) || !ktime_after(alarm_time, now))
> +			alarm_time = down;
> +		else
> +			alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);

[Severity: High]
Does 60 * NSEC_PER_SEC overflow a 32-bit signed long on 32-bit
architectures?

NSEC_PER_SEC is 1000000000L and ktime_add_ns() is a plain macro addition,
so 60 * NSEC_PER_SEC is evaluated in signed long arithmetic before promotion
to ktime_t.  On 32-bit builds this triggers a -Woverflow build warning (or
error under CONFIG_WERROR) and wraps to a negative offset at runtime,
leaving alarm_time in the past.  The ktime_sub_ns() call above casts to u64
to avoid this; using 60ULL * NSEC_PER_SEC here would match it.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261110-rtc-minute-alarm-v1-1-be9d7c5227a1@gmail.com?part=1

Re: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[Alexandre Belloni]

On 10/11/2026 07:13:47+0000, Arun Iyengar via B4 Relay wrote:
> From: Arun Iyengar <aiyengar1990@gmail.com>
> 
> For an RTC with RTC_FEATURE_ALARM_RES_MINUTE, rtc_set_alarm() rounds the
> alarm down to the start of its minute, so that it never fires late. When
> the alarm is less than a minute away and in the same minute as the
> current time, the rounded time is already in the past.
> rtc_timer_enqueue() then gets -ETIME from __rtc_set_alarm() and expires
> the timer from the irq work right away, and the hardware alarm is never
> armed.

This is the expected behaviour, userspace has to handle the error
properly: the alarm can't be set, set it for a later time or stop there
and don't suspend.

> 
> That loses every short wake alarm. With a HYM8563, which has minute
> resolution, "rtcwake -m freeze -s 15" sets an alarm that expires in
> software before the system suspends, so nothing wakes it: with the
> alarm set at 17:55:00 for 17:55:17, /proc/driver/rtc reads back
> "alrm_time: 17:55:00, alarm_IRQ: no", and the system stays in
> suspend-to-idle until something else wakes it.
> 
> Round the alarm up to the next minute when rounding down would put an
> alarm that is still in the future into the past. It then fires at most
> 59 seconds late instead of never. Alarms that are already in the past,
> or that round down to a time still in the future, are handled as before.

Never round up, you are going to miss deadlines.


-- 
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Re: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[Arun Iyengar]

Hi Alexandre,

> This is the expected behaviour, userspace has to handle the error
> properly: the alarm can't be set, set it for a later time or stop there
> and don't suspend.

OK. One detail: userspace gets no error in this case.
rtc_timer_enqueue() turns the -ETIME from __rtc_set_alarm() into an
immediate expiry and returns 0, so RTC_WKALM_SET and the wakealarm
sysfs file succeed. What userspace gets is an RTC_AF event right away
and a disabled alarm from RTC_WKALM_RD.

I checked that on the HYM8563 without my patch: RTC_WKALM_SET for 15 s
ahead in the current minute returned 0 in 5 of 5 tries. rtcwake only
checks that return value, so "rtcwake -m freeze -s 15" suspended with
no alarm armed and did not wake up.

> Never round up, you are going to miss deadlines.

Understood, please drop the patch. I'll fix rtcwake instead: read the
alarm back after setting it and not suspend when it has already
expired.

Would you also want rtc_set_alarm() to return -ETIME when the rounding
moves an alarm that is still in the future into the past? If the
immediate RTC_AF is the intended signal, I'll leave the kernel as it
is.

Thanks,
Arun

On Mon, Nov 9, 2026 at 11:48 PM Alexandre Belloni
<alexandre.belloni@bootlin.com> wrote:
>
> On 10/11/2026 07:13:47+0000, Arun Iyengar via B4 Relay wrote:
> > From: Arun Iyengar <aiyengar1990@gmail.com>
> >
> > For an RTC with RTC_FEATURE_ALARM_RES_MINUTE, rtc_set_alarm() rounds the
> > alarm down to the start of its minute, so that it never fires late. When
> > the alarm is less than a minute away and in the same minute as the
> > current time, the rounded time is already in the past.
> > rtc_timer_enqueue() then gets -ETIME from __rtc_set_alarm() and expires
> > the timer from the irq work right away, and the hardware alarm is never
> > armed.
>
> This is the expected behaviour, userspace has to handle the error
> properly: the alarm can't be set, set it for a later time or stop there
> and don't suspend.
>
> >
> > That loses every short wake alarm. With a HYM8563, which has minute
> > resolution, "rtcwake -m freeze -s 15" sets an alarm that expires in
> > software before the system suspends, so nothing wakes it: with the
> > alarm set at 17:55:00 for 17:55:17, /proc/driver/rtc reads back
> > "alrm_time: 17:55:00, alarm_IRQ: no", and the system stays in
> > suspend-to-idle until something else wakes it.
> >
> > Round the alarm up to the next minute when rounding down would put an
> > alarm that is still in the future into the past. It then fires at most
> > 59 seconds late instead of never. Alarms that are already in the past,
> > or that round down to a time still in the future, are handled as before.
>
> Never round up, you are going to miss deadlines.
>
>
> --
> Alexandre Belloni, co-owner and COO, Bootlin
> Embedded Linux and Kernel engineering
> https://bootlin.com

Re: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[kernel test robot]

Hi Arun,

kernel test robot noticed the following build warnings:

[auto build test WARNING on 7485659ecb81d5fce3a3edf2cae9fcb29c17b1cd]

url:    https://github.com/intel-lab-lkp/linux/commits/Arun-Iyengar-via-B4-Relay/rtc-Don-t-lose-a-minute-resolution-alarm-that-rounds-down-into-the-past/20261110-071347
base:   7485659ecb81d5fce3a3edf2cae9fcb29c17b1cd
patch link:    https://lore.kernel.org/r/20261110-rtc-minute-alarm-v1-1-be9d7c5227a1%40gmail.com
patch subject: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past
config: parisc-allnoconfig (https://download.01.org/0day-ci/archive/20261110/202611101839.fRQuN29G-lkp@intel.com/config)
compiler: hppa-linux-gcc (GCC) 16.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20261110/202611101839.fRQuN29G-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202611101839.fRQuN29G-lkp@intel.com/

All warnings (new ones prefixed by >>):

   In file included from include/linux/timer.h:6,
                    from include/linux/uprobes.h:18,
                    from include/linux/mm_types.h:16,
                    from include/linux/mmzone.h:22,
                    from include/linux/topology.h:34,
                    from include/linux/irq.h:19,
                    from arch/parisc/include/asm/hardirq.h:13,
                    from include/linux/hardirq.h:11,
                    from include/linux/interrupt.h:11,
                    from include/linux/rtc.h:18,
                    from drivers/rtc/interface.c:11:
   drivers/rtc/interface.c: In function 'rtc_set_alarm':
>> drivers/rtc/interface.c:562:60: warning: integer overflow in expression of type 'long int' results in '-129542144' [-Woverflow]
     562 |                         alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);
         |                                                            ^
   include/linux/ktime.h:60:50: note: in definition of macro 'ktime_add_ns'
      60 | #define ktime_add_ns(kt, nsval)         ((kt) + (nsval))
         |                                                  ^~~~~


vim +562 drivers/rtc/interface.c

   512	
   513	int rtc_set_alarm(struct rtc_device *rtc, struct rtc_wkalrm *alarm)
   514	{
   515		ktime_t alarm_time;
   516		int err;
   517	
   518		if (!rtc->ops)
   519			return -ENODEV;
   520		else if (!test_bit(RTC_FEATURE_ALARM, rtc->features))
   521			return -EINVAL;
   522	
   523		err = rtc_valid_tm(&alarm->time);
   524		if (err != 0)
   525			return err;
   526	
   527		err = rtc_valid_range(rtc, &alarm->time);
   528		if (err)
   529			return err;
   530	
   531		err = mutex_lock_interruptible(&rtc->ops_lock);
   532		if (err)
   533			return err;
   534		if (rtc->aie_timer.enabled)
   535			rtc_timer_remove(rtc, &rtc->aie_timer);
   536	
   537		alarm_time = rtc_tm_to_ktime(alarm->time);
   538		if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features) &&
   539		    alarm->time.tm_sec) {
   540			ktime_t down = ktime_sub_ns(alarm_time,
   541						    (u64)alarm->time.tm_sec * NSEC_PER_SEC);
   542			struct rtc_time tm;
   543			ktime_t now;
   544	
   545			err = __rtc_read_time(rtc, &tm);
   546			if (err)
   547				goto unlock;
   548			now = rtc_tm_to_ktime(tm);
   549	
   550			/*
   551			 * Round down so we never miss a deadline. If that puts an alarm
   552			 * that is still in the future into the past, the timer would
   553			 * expire right away and the hardware alarm would never be
   554			 * armed: whoever relies on it, e.g. to wake the system from
   555			 * suspend, would wait forever. Round up to the next minute
   556			 * instead. An alarm already in the past is left alone, it is
   557			 * handled by __rtc_set_alarm.
   558			 */
   559			if (ktime_after(down, now) || !ktime_after(alarm_time, now))
   560				alarm_time = down;
   561			else
 > 562				alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);
   563		}
   564	
   565		rtc->aie_timer.node.expires = alarm_time;
   566		rtc->aie_timer.period = 0;
   567		if (alarm->enabled)
   568			err = rtc_timer_enqueue(rtc, &rtc->aie_timer);
   569	
   570	unlock:
   571		mutex_unlock(&rtc->ops_lock);
   572	
   573		return err;
   574	}
   575	EXPORT_SYMBOL_GPL(rtc_set_alarm);
   576	

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past

[kernel test robot]

Hi Arun,

kernel test robot noticed the following build warnings:

[auto build test WARNING on 7485659ecb81d5fce3a3edf2cae9fcb29c17b1cd]

url:    https://github.com/intel-lab-lkp/linux/commits/Arun-Iyengar-via-B4-Relay/rtc-Don-t-lose-a-minute-resolution-alarm-that-rounds-down-into-the-past/20261110-071347
base:   7485659ecb81d5fce3a3edf2cae9fcb29c17b1cd
patch link:    https://lore.kernel.org/r/20261110-rtc-minute-alarm-v1-1-be9d7c5227a1%40gmail.com
patch subject: [PATCH] rtc: Don't lose a minute-resolution alarm that rounds down into the past
config: i386-defconfig (https://download.01.org/0day-ci/archive/20261110/202611102202.ZY3wJ35v-lkp@intel.com/config)
compiler: clang version 22.1.8 (https://github.com/llvm/llvm-project ca7933e47d3a3451d81e72ac174dcb5aa28b59d1)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20261110/202611102202.ZY3wJ35v-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202611102202.ZY3wJ35v-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> drivers/rtc/interface.c:562:39: warning: overflow in expression; result is -129'542'144 with type 'long' [-Winteger-overflow]
     562 |                         alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);
         |                                      ~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
   include/linux/ktime.h:60:43: note: expanded from macro 'ktime_add_ns'
      60 | #define ktime_add_ns(kt, nsval)         ((kt) + (nsval))
         |                                                  ^~~~~
   1 warning generated.


vim +562 drivers/rtc/interface.c

   512	
   513	int rtc_set_alarm(struct rtc_device *rtc, struct rtc_wkalrm *alarm)
   514	{
   515		ktime_t alarm_time;
   516		int err;
   517	
   518		if (!rtc->ops)
   519			return -ENODEV;
   520		else if (!test_bit(RTC_FEATURE_ALARM, rtc->features))
   521			return -EINVAL;
   522	
   523		err = rtc_valid_tm(&alarm->time);
   524		if (err != 0)
   525			return err;
   526	
   527		err = rtc_valid_range(rtc, &alarm->time);
   528		if (err)
   529			return err;
   530	
   531		err = mutex_lock_interruptible(&rtc->ops_lock);
   532		if (err)
   533			return err;
   534		if (rtc->aie_timer.enabled)
   535			rtc_timer_remove(rtc, &rtc->aie_timer);
   536	
   537		alarm_time = rtc_tm_to_ktime(alarm->time);
   538		if (test_bit(RTC_FEATURE_ALARM_RES_MINUTE, rtc->features) &&
   539		    alarm->time.tm_sec) {
   540			ktime_t down = ktime_sub_ns(alarm_time,
   541						    (u64)alarm->time.tm_sec * NSEC_PER_SEC);
   542			struct rtc_time tm;
   543			ktime_t now;
   544	
   545			err = __rtc_read_time(rtc, &tm);
   546			if (err)
   547				goto unlock;
   548			now = rtc_tm_to_ktime(tm);
   549	
   550			/*
   551			 * Round down so we never miss a deadline. If that puts an alarm
   552			 * that is still in the future into the past, the timer would
   553			 * expire right away and the hardware alarm would never be
   554			 * armed: whoever relies on it, e.g. to wake the system from
   555			 * suspend, would wait forever. Round up to the next minute
   556			 * instead. An alarm already in the past is left alone, it is
   557			 * handled by __rtc_set_alarm.
   558			 */
   559			if (ktime_after(down, now) || !ktime_after(alarm_time, now))
   560				alarm_time = down;
   561			else
 > 562				alarm_time = ktime_add_ns(down, 60 * NSEC_PER_SEC);
   563		}
   564	
   565		rtc->aie_timer.node.expires = alarm_time;
   566		rtc->aie_timer.period = 0;
   567		if (alarm->enabled)
   568			err = rtc_timer_enqueue(rtc, &rtc->aie_timer);
   569	
   570	unlock:
   571		mutex_unlock(&rtc->ops_lock);
   572	
   573		return err;
   574	}
   575	EXPORT_SYMBOL_GPL(rtc_set_alarm);
   576	

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki