Setting System Clock From External RTC At Startup

[ebs27]

Hi,

I have a Jetson Orin Nano Developer Kit Running Ubuntu 22.04 and Jetpack 6.

I am trying to set my system clock to the hardware clock time using an external RTC over I2C (ds3231 from Adafruit Adafruit PiRTC - Precise DS3231 Real Time Clock for Raspberry Pi : ID 4282 : Adafruit Industries, Unique & fun DIY electronics and kits). I tried using the built-in clock on the Jetson, but had issues soldering the surface mount battery connector and resistor to the board…

Anyways, here is what I have so far:

• I have the external RTC setup as rtc2 and connected to the Jetson I2C pins (rtc0 and rtc1 are internal to the Jetson)
• The external RTC is successfully maintaining its clock when the Jetson is powered down
• I am able to manually set the system clock to rtc2’s hardware using the command "sudo hwclock -s -f /dev/rtc2
• I have disabled NTP/automatic time updating from the internet

I would like to have the system clock automatically update to the external rtc2 clock at startup. I have tried a few things. So far, the Jetson always resets to the rtc0 clock after reboot.

I have tried running the sudo hwclock -s -f /dev/rtc2 command in crontab (both as local and sudo). I’ve confirmed my cron service is running and the command runs on its own. I have also tried modifying to the /etc/init.d/hwclock.sh.

Is there a way to change the rtc clock configurations so that my external clock is rtc0? Or is there another file I need to change to have it use rtc2 as the source?

Many thanks!

[ebs27]

#!/bin/sh
timedatectl >> Desktop/date.txt
echo ds3231 0x68 | sudo tee /sys/class/i2c-dev/i2c-7/device/new_device
sudo hwclock --hctosys -f /dev/rtc2

FYI this is the script that I am running to set the system clock to the external RTC clock. It works if I just run from a command line once the system is already boot. Any reason why it doesn’t work when I try and run at startup (thru cron)?

[JerryChang]

hello ebs27,

here’s kernel driver for setting date and reading it back.
please check whether it’s reporting any errors.
$public_sources/kernel_src/kernel/nvidia-oot/drivers/rtc/nvvrs-pseq-rtc.c

static int nvvrs_rtc_read_time(struct device *dev, struct rtc_time *tm)
{
...
        ret = i2c_smbus_read_byte_data(client, info->drv_data->map[RTC_T2]);
        if (ret < 0) {
                dev_err(info->dev, "Failed to read time reg:0x%x ret:(%d)\n",
                        info->drv_data->map[RTC_T2], ret);
                goto out;
        }   

static int nvvrs_rtc_set_time(struct device *dev, struct rtc_time *tm)
{
...
        ret = i2c_smbus_write_byte_data(client, info->drv_data->map[RTC_T2], time[2]);
        if (ret < 0) {
                dev_err(info->dev, "Failed to write time reg: 0x%x ret:(%d)\n",
                        info->drv_data->map[RTC_T2], ret);
                goto out;
        }

[ebs27]

I’ve got it work! I’m now able to set the system clock from my external rtc (rtc2) via a start-up script. I also tried this as rc.local startup script instead of cron, since I read somewhere in NVIDIA FAQs that that is how you should handle startup scripts on Jetsons

Old script

#!/bin/sh
echo ds3231 0x68 | sudo tee /sys/class/i2c-dev/i2c-7/device/new_device
sudo hwclock --hctosys -f /dev/rtc2 --verbose

I dug a little deeper and found that the first line (echo…) was working properly and creating the device path. I added the --verbose flag and looked at the system status message after booting the Jetson and found that the second line (hwclock…) was failing because “No usable clock interface found.” I thought this was strange because it has root access and I can run that command manually after the cpu had booted. Maybe something wasn’t fully initialized at that point?

New Script

#!/bin/sh
echo ds3231 0x68 | sudo tee /sys/class/i2c-dev/i2c-7/device/new_device
sleep 1
sudo hwclock --hctosys -f /dev/rtc2 --verbose

The solution was to add a sleep (used an arbitrary 1 second, it could probably be lower?) between the two lines. This works.

I haven’t gone back and tried with cron, but I imagine it would work too and that moving the commands to rc.local might not have been necessary. There might be a more “proper” solution for this, but it works for my situation and it’s simple so I’m marking it as a solution.

[JerryChang]

hello ebs27,

thanks for sharing, glad to know it works.

@ebs27

used an arbitrary 1 second, it could probably be lower?

please refer to service config, for instance, /etc/systemd/system/nv.service
it should works with After commands, which waiting for device to initial completely.