[systemd-devel] hwclock

Kay Sievers kay.sievers at vrfy.org
Sun Jan 23 05:28:48 PST 2011


On Sun, Jan 23, 2011 at 00:01, Tom Gundersen <teg at jklm.no> wrote:
> Further to my previous mail about hwclock, I'm struggling to get my
> head around what is the "proper" way to handle hwclock's adjfile.
>
> These are my assumptions:
>
> /etc is in principle mounted read-only (not yet possible, but that's
> the goal, right?)
> /var is in principle not available during early boot
> the drift of the hardware clock is updated at regular intervals and
> cannot be on a read-only filesystem (but it is ok that we don't know
> the drift at early boot (is this true?))
> we need to know whether or not the hardware clock is in utc during
> early boot (is this actually required? if not, could someone point me
> to a discussion/documentation?)
>
> If all of this is correct, it seems to me that the only solution
> (without changing hwclock) is to ignore the utc/localtime part of
> adjfile and have a config file in /etc that stores whether or not we
> use utc (resp., localtime) and always invoke hwclock with
> "--adjfile=/var/hwclock/adjfile --utc" (resp.,
> "--adjfile=/var/hwclock/adjfile --localtime").
>
> I realise that if someone oblivious to this policy were to call
> hwclock without --adjfile, havoc would ensue, but this is the best I
> could come up with.
>
> I'd be happy to prepare a patch, if someone can confirm that my
> approach is sound.

All that adjfile hackery should probably just be left alone. If you
need proper time, use ntp, if not, leave the rtc as it is. I really
doubt its usefulness these days.

Is that stuff really useful for you, or do you just port-over old stuff?

Kay


More information about the systemd-devel mailing list