connman + busybox-hwclock startup failure

connman + busybox-hwclock startup failure

[Chris Elledge]

[-- Attachment #1: Type: text/plain, Size: 978 bytes --]

I've encountered an issue when using a system that has both connman and the
hwclock.sh script enabled during startup. The default combined startup
order for these two services is wrong.

Currently connman starts at 05, and hwclock.sh is at 20 (default for
update-rc.d). This means that we have a situation where connman starts up
and obtains a correct NTP time, then hwclock.sh clobbers the system time
with whatever is in the RTC even if it is way in the past. The big issue
caused by that is connman schedules its next NTP update at an absolute time
of (current time + 1024s). Once the hwclock.sh script knocks the system
time way into the past, the NTP update will not run for a potentially very
long time.

I would suggest adding a line to poky/meta/recipes-core/busybox.inc to move
hwclock.sh earlier.
For example:
INITSCRIPT_PARAMS:${PN}-hwclock = "start 04 2 3 4 5 . stop 20 0 1 6"
or
INITSCRIPT_PARAMS:${PN}-hwclock = "start 40 S . stop 20 0 1 6"

Thanks,
Chris Elledge

[-- Attachment #2: Type: text/html, Size: 1187 bytes --]

Re: [yocto] connman + busybox-hwclock startup failure

[Alexander Kanavin]

Can you submit a patch for this please? The mailing list is openembedded-core@.

Alex

On Fri, 31 Mar 2023 at 00:05, Chris Elledge <celledge@siteworx.io> wrote:
>
> I've encountered an issue when using a system that has both connman and the hwclock.sh script enabled during startup. The default combined startup order for these two services is wrong.
>
> Currently connman starts at 05, and hwclock.sh is at 20 (default for update-rc.d). This means that we have a situation where connman starts up and obtains a correct NTP time, then hwclock.sh clobbers the system time with whatever is in the RTC even if it is way in the past. The big issue caused by that is connman schedules its next NTP update at an absolute time of (current time + 1024s). Once the hwclock.sh script knocks the system time way into the past, the NTP update will not run for a potentially very long time.
>
> I would suggest adding a line to poky/meta/recipes-core/busybox.inc to move hwclock.sh earlier.
> For example:
> INITSCRIPT_PARAMS:${PN}-hwclock = "start 04 2 3 4 5 . stop 20 0 1 6"
> or
> INITSCRIPT_PARAMS:${PN}-hwclock = "start 40 S . stop 20 0 1 6"
>
> Thanks,
> Chris Elledge
>
> -=-=-=-=-=-=-=-=-=-=-=-
> Links: You receive all messages sent to this group.
> View/Reply Online (#59555): https://lists.yoctoproject.org/g/yocto/message/59555
> Mute This Topic: https://lists.yoctoproject.org/mt/97961702/1686489
> Group Owner: yocto+owner@lists.yoctoproject.org
> Unsubscribe: https://lists.yoctoproject.org/g/yocto/unsub [alex.kanavin@gmail.com]
> -=-=-=-=-=-=-=-=-=-=-=-
>

Re: [yocto] connman + busybox-hwclock startup failure

[Mike Looijmans]

A bit late, but might be useful to others.

The Linux kernel has options to set the RTC time at boot. This will happen as 
soon as the RTC probes, which is often before the rootfs mounts.

The kernel also has an option to synchronize the RTC to NTP time (userspace 
provides the actual NTP) when available.

Combined this completely replaces the hwclock functionality. In fact, I found 
that I had to bbappend the busybox recipeand remove the hwclock scripts, they 
actually interfere with proper operation:
- At boot, the RTC clock is copied twice
- At shutdown the RTC is set. But most RTC clocks have more accurate time 
keeping (e.g. temperature compensated crystal) than what the system has, so 
this usually makes things worse. Better to set the RTC time from userspace and 
leave it alone.
- Most RTCs have only "second" resolution and setting it too often makes it 
drift more.



On 31-03-2023 00:04, Chris Elledge via lists.yoctoproject.org wrote:
> I've encountered an issue when using a system that has both connman and the 
> hwclock.sh script enabled during startup. The default combined startup order 
> for these two services is wrong.
> 
> Currently connman starts at 05, and hwclock.sh is at 20 (default for 
> update-rc.d). This means that we have a situation where connman starts up and 
> obtains a correct NTP time, then hwclock.sh clobbers the system time with 
> whatever is in the RTC even if it is way in the past. The big issue caused by 
> that is connman schedules its next NTP update at an absolute time of (current 
> time + 1024s). Once the hwclock.sh script knocks the system time way into the 
> past, the NTP update will not run for a potentially very long time.
> 
> I would suggest adding a line to poky/meta/recipes-core/busybox.inc to move 
> hwclock.sh earlier.
> For example:
> INITSCRIPT_PARAMS:${PN}-hwclock = "start 04 2 3 4 5 . stop 20 0 1 6"
> or
> INITSCRIPT_PARAMS:${PN}-hwclock = "start 40 S . stop 20 0 1 6"
> 
> Thanks,
> Chris Elledge
> 
> 

-- 
Mike Looijmans
System Expert

TOPIC Embedded Products B.V.
Materiaalweg 4, 5681 RJ Best
The Netherlands

T: +31 (0) 499 33 69 69
E: mike.looijmans@topic.nl
W: www.topic.nl

Re: [yocto] connman + busybox-hwclock startup failure

[Chris Elledge]

[-- Attachment #1: Type: text/plain, Size: 201 bytes --]

I accidentally omitted the final "." in my examples.

INITSCRIPT_PARAMS:${PN}-hwclock = "start 04 2 3 4 5 . stop 20 0 1 6 ."
or
INITSCRIPT_PARAMS:${PN}-hwclock = "start 40 S . stop 20 0 1 6 ."

-Chris

[-- Attachment #2: Type: text/html, Size: 373 bytes --]