hwclock, rtcwake, /etc/adjtime

[Paul Fox]

executive summary:  we need to make sure that /etc/adjtime, as
shipped, has a third line that says "UTC".  alternatively, we
should add such a line on first boot.

having already written all this out for myself, to be sure i
understand it, i'll share:

    - the third line in /etc/adjtime specifies whether the
	RTC is maintained in localtime or UTC.  most unix
	systems maintain the RTC in UTC.

    - hwclock needs (and maintains) the third line, because
	hwclock can set the system time from the RTC, and
	vice-versa.  if there's no third line, hwclock assumes
	the RTC is maintained in localtime.  (this surprises me,
	but i think it's always been true.)  hwclock will 
	add a third line if it's missing, and the added line
	will say "LOCAL".

    - rtcwake needs the third line, because it uses the RTC to
	wake up the system at a given time in the future.  if
	rtcwake finds no third line in /etc/adjtime, it assumes UTC.
	please note the irony that hwclock and rtcwake are both
	provided by the same package (util-linux-ng).

    - rtcwake has a bug with regard to daylight savings time
	which only shows up if it thinks the RTC is in maintained
	in localtime.  the bug causes wakeup times to be wrong by
	at least an hour.

    - when an XO boots for the first time, the /etc/adjtime file
	provided by the initscripts package has no third line.

    - hwclock is run by /etc/init.d/halt when the system is
	shutdown cleanly, at which point /etc/adjtime will
	have it's third line created, to say "LOCAL".

    - XO laptops have their RTC set to UTC at the factory.

the upshot of all this is that on initial boot of an XO,
rtcwake will work fine, and after the first shutdown, it will
not.  i'm sure there are also implications for the system
time and date, and the system timezone.  but i haven't thought
about those issues much.

anyway, i think it can all be fixed by shipping a complete
/etc/adjtime file.

paul
=---------------------
 paul fox, pgf(a)laptop.org

[Daniel Drake]

2009/8/22 Paul Fox <pgf(a)laptop.org&gt;:
>  the upshot of all this is that on initial boot of an XO,
>  rtcwake will work fine, and after the first shutdown, it will
>  not.  i'm sure there are also implications for the system
>  time and date, and the system timezone.  but i haven't thought
>  about those issues much. 
I'll bet that this is already fixed for the next build with my
kickstart changes. adjtime will be empty on first boot, and then
populated with "UTC" during first shutdown, and if I understand your
email correctly then that will produce the desired behaviour.

Daniel

[Paul Fox]

daniel wrote:
>  2009/8/22 Paul Fox <pgf(a)laptop.org&gt;:
>  > the upshot of all this is that on initial boot of an XO,
>  > rtcwake will work fine, and after the first shutdown, it will
>  > not. �i'm sure there are also implications for the system
>  > time and date, and the system timezone. �but i haven't thought
>  > about those issues much.
>  
>  I'll bet that this is already fixed for the next build with my
>  kickstart changes. adjtime will be empty on first boot, and then
>  populated with "UTC" during first shutdown, and if I understand your
>  email correctly then that will produce the desired behaviour. 
okay.  you had said that kickstart creates /etc/sysconfig/clock,
and that that's what you'd fixed.  mtd's distro has "UTC=True" in
that file, so i figured he already had your fix.  there must be
more to it i guess.  oh -- unless soas-on-xo1 doesn't use kickstart,
and more needs to happen if it doesn't.

paul
=---------------------
 paul fox, pgf(a)laptop.org

[Daniel Drake]

2009/8/22 Paul Fox <pgf(a)laptop.org&gt;:
>  okay.  you had said that kickstart creates /etc/sysconfig/clock,
>  and that that's what you'd fixed.  mtd's distro has "UTC=True" in
>  that file, so i figured he already had your fix.  there must be
>  more to it i guess.  oh -- unless soas-on-xo1 doesn't use kickstart,
>  and more needs to happen if it doesn't. 
ok, maybe it's not the complete fix then.

Here's how I would expect things to work:

kickstart puts UTC=True in /etc/sysconfig/clock
this makes fedora init scripts always call hwclock with --utc
and this option makes hwclock write "UTC" in /etc/adjtime

Perhaps my understanding is flawed/incomplete, or there is a bug somewhere.

Daniel

[Paul Fox]

daniel wrote:
>  2009/8/22 Paul Fox <pgf(a)laptop.org&gt;:
>  > okay. �you had said that kickstart creates /etc/sysconfig/clock,
>  > and that that's what you'd fixed. �mtd's distro has "UTC=True"
> in
>  > that file, so i figured he already had your fix. �there must be
>  > more to it i guess. �oh -- unless soas-on-xo1 doesn't use kickstart,
>  > and more needs to happen if it doesn't.
>  
>  ok, maybe it's not the complete fix then.
>  
>  Here's how I would expect things to work:
>  
>  kickstart puts UTC=True in /etc/sysconfig/clock
>  this makes fedora init scripts always call hwclock with --utc
>  and this option makes hwclock write "UTC" in /etc/adjtime 
that's certainly a reasonable expectation.  :-)

but i find no reference to /etc/sysconfig/clock in any init.d
script, either on the SoaS release or on learn, which is running
a 10.91 rawhide.  there's no facility in either the halt or
ntpdate scripts for hwclocks commandline arguments to be
influenced, so it must be that /etc/adjtime is normally modified
by some other startup or installation script.

>  
>  Perhaps my understanding is flawed/incomplete, or there is a bug somewhere. 
makes two of us.

paul
=---------------------
 paul fox, pgf(a)laptop.org

[Martin Dengler]

On Sat, Aug 22, 2009 at 10:41:14AM +0545, Daniel Drake wrote:
>  2009/8/22 Paul Fox <pgf(a)laptop.org&gt;:
>  > okay.  you had said that kickstart creates /etc/sysconfig/clock,
>  > and that that's what you'd fixed.  mtd's distro has "UTC=True"
> in
>  > that file, so i figured he already had your fix. 
Yes, I/soas-on-xo1 does have the fix.

>  > oh -- unless soas-on-xo1 doesn't use kickstart, and more
> needs to
>  > happen if it doesn't. 
soas-on-xo1 does use kickstart.

>  ok, maybe it's not the complete fix then. 
It's not, it appears, due to the below.  That being said, I've just:

- created a new image and booted it fine
- confirmed /etc/adjtime has neither LOCAL nor UTC
- tested rtcwake (works fine)
- confirmed /etc/sysconfig/clock has "UTC=True"
- rebooted
- confirmed /etc/adjtime now has LOCAL in it (see below)
- tested rtcwake (works fine)

so go figure.


>  Here's how I would expect things to work:
>  
>  kickstart puts UTC=True in /etc/sysconfig/clock 
Yes, this happens

>  this makes fedora init scripts always call hwclock with --utc 
This is the problem.  /etc/init.d/halt just does:

/sbin/hwclock --systohc

...which is documented[1] to default to localtime in these
circumstances.

Also, 'grep UTC /etc/init.d/*' turns up nothing.

>  and this option makes hwclock write "UTC" in /etc/adjtime
...so this never happens.

>  Perhaps my understanding is flawed/incomplete, or there is a bug
>  somewhere. 
Indeed.  At the very least, I'm not sure what the point of
/etc/sysconfig/clock's UTC=true setting is.

>  Daniel 
Martin


1.
http://git.kernel.org/?p=utils/util-linux-ng/util-linux-ng.git;a=blob;f=h...