The VRTC alarm register can be programmed with an amount of seconds
after which the SoC will be woken up by the VRTC timer again. We are
already converting the alarm time from meson_vrtc_set_alarm() to
"seconds since 1970". This means we also need to use "seconds since
1970" for the current time.
This fixes a problem where setting the alarm to one minute in the future
results in the firmware (which handles wakeup) to output (on the serial
console) that the system will be woken up in billions of seconds.
ktime_get_raw_ts64() returns the time since boot, not since 1970. Switch
to ktime_get_real_ts64() to fix the calculation of the alarm time and to
make the SoC wake up at the specified date/time. Also the firmware
(which manages suspend) now prints either 59 or 60 seconds until wakeup
(depending on how long it takes for the system to enter suspend).
Fixes: 6ef35398e827 ("rtc: Add Amlogic Virtual Wake RTC")
Signed-off-by: Martin Blumenstingl <[email protected]>
---
drivers/rtc/rtc-meson-vrtc.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/rtc/rtc-meson-vrtc.c b/drivers/rtc/rtc-meson-vrtc.c
index 1463c8621561..648fa362ec44 100644
--- a/drivers/rtc/rtc-meson-vrtc.c
+++ b/drivers/rtc/rtc-meson-vrtc.c
@@ -23,7 +23,7 @@ static int meson_vrtc_read_time(struct device *dev, struct rtc_time *tm)
struct timespec64 time;
dev_dbg(dev, "%s\n", __func__);
- ktime_get_raw_ts64(&time);
+ ktime_get_real_ts64(&time);
rtc_time64_to_tm(time.tv_sec, tm);
return 0;
@@ -96,7 +96,7 @@ static int __maybe_unused meson_vrtc_suspend(struct device *dev)
long alarm_secs;
struct timespec64 time;
- ktime_get_raw_ts64(&time);
+ ktime_get_real_ts64(&time);
local_time = time.tv_sec;
dev_dbg(dev, "alarm_time = %lus, local_time=%lus\n",
--
2.40.0
Martin Blumenstingl <[email protected]> writes:
> The VRTC alarm register can be programmed with an amount of seconds
> after which the SoC will be woken up by the VRTC timer again. We are
> already converting the alarm time from meson_vrtc_set_alarm() to
> "seconds since 1970". This means we also need to use "seconds since
> 1970" for the current time.
>
> This fixes a problem where setting the alarm to one minute in the future
> results in the firmware (which handles wakeup) to output (on the serial
> console) that the system will be woken up in billions of seconds.
> ktime_get_raw_ts64() returns the time since boot, not since 1970. Switch
> to ktime_get_real_ts64() to fix the calculation of the alarm time and to
> make the SoC wake up at the specified date/time. Also the firmware
> (which manages suspend) now prints either 59 or 60 seconds until wakeup
> (depending on how long it takes for the system to enter suspend).
>
> Fixes: 6ef35398e827 ("rtc: Add Amlogic Virtual Wake RTC")
> Signed-off-by: Martin Blumenstingl <[email protected]>
Reviewed-by: Kevin Hilman <[email protected]>
On 20/03/2023 22:21, Martin Blumenstingl wrote:
> The VRTC alarm register can be programmed with an amount of seconds
> after which the SoC will be woken up by the VRTC timer again. We are
> already converting the alarm time from meson_vrtc_set_alarm() to
> "seconds since 1970". This means we also need to use "seconds since
> 1970" for the current time.
>
> This fixes a problem where setting the alarm to one minute in the future
> results in the firmware (which handles wakeup) to output (on the serial
> console) that the system will be woken up in billions of seconds.
> ktime_get_raw_ts64() returns the time since boot, not since 1970. Switch
> to ktime_get_real_ts64() to fix the calculation of the alarm time and to
> make the SoC wake up at the specified date/time. Also the firmware
> (which manages suspend) now prints either 59 or 60 seconds until wakeup
> (depending on how long it takes for the system to enter suspend).
>
> Fixes: 6ef35398e827 ("rtc: Add Amlogic Virtual Wake RTC")
> Signed-off-by: Martin Blumenstingl <[email protected]>
> ---
> drivers/rtc/rtc-meson-vrtc.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/rtc/rtc-meson-vrtc.c b/drivers/rtc/rtc-meson-vrtc.c
> index 1463c8621561..648fa362ec44 100644
> --- a/drivers/rtc/rtc-meson-vrtc.c
> +++ b/drivers/rtc/rtc-meson-vrtc.c
> @@ -23,7 +23,7 @@ static int meson_vrtc_read_time(struct device *dev, struct rtc_time *tm)
> struct timespec64 time;
>
> dev_dbg(dev, "%s\n", __func__);
> - ktime_get_raw_ts64(&time);
> + ktime_get_real_ts64(&time);
> rtc_time64_to_tm(time.tv_sec, tm);
>
> return 0;
> @@ -96,7 +96,7 @@ static int __maybe_unused meson_vrtc_suspend(struct device *dev)
> long alarm_secs;
> struct timespec64 time;
>
> - ktime_get_raw_ts64(&time);
> + ktime_get_real_ts64(&time);
> local_time = time.tv_sec;
>
> dev_dbg(dev, "alarm_time = %lus, local_time=%lus\n",
Thx for the fix!
Reviewed-by: Neil Armstrong <[email protected]>
On Mon, 20 Mar 2023 22:21:42 +0100, Martin Blumenstingl wrote:
> The VRTC alarm register can be programmed with an amount of seconds
> after which the SoC will be woken up by the VRTC timer again. We are
> already converting the alarm time from meson_vrtc_set_alarm() to
> "seconds since 1970". This means we also need to use "seconds since
> 1970" for the current time.
>
> This fixes a problem where setting the alarm to one minute in the future
> results in the firmware (which handles wakeup) to output (on the serial
> console) that the system will be woken up in billions of seconds.
> ktime_get_raw_ts64() returns the time since boot, not since 1970. Switch
> to ktime_get_real_ts64() to fix the calculation of the alarm time and to
> make the SoC wake up at the specified date/time. Also the firmware
> (which manages suspend) now prints either 59 or 60 seconds until wakeup
> (depending on how long it takes for the system to enter suspend).
>
> [...]
Applied, thanks!
[1/1] rtc: meson-vrtc: Use ktime_get_real_ts64() to get the current time
commit: 0e6255fa3f649170da6bd1a544680589cfae1131
Best regards,
--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com