[F][PATCH 0/1] rtc: class: support hctosys from modular RTC drivers

[Jesse Sung]

On Thu, Nov 25, 2021 at 12:00 AM Dimitri John Ledkov
<dimitri.ledkov at canonical.com> wrote:
>
> But doesn't this break assumptions that systemd makes? I.e. it would
> sync it's time from the timesync state file, and then loading rtc
> module would trump this.
>
> This change in v5.7 kernel has caused breakage. Does introducing this
> in v5.4 kernel going to break UC20 machines which sync a better/secure
> time in the initrd; and then will get broken time from the rtc module
> because it happens to not be battery backed. Leading to inability to
> snap refresh?

IMHO a proper solution for this is to have a way to add a list of
modules when we use ubuntu-core-initramfs to generate the initrd.
Something like /etc/initramfs-tools/modules in the classic image. For
example:
ubuntu-core-initram --extra-kernel-modules <mod1>,<mod2>,<mod3>
So on one hand we don't need to make every RTC driver built-in, and on
the other hand hctosys happens before systemd tries to access system
time hence it acts like a built-in driver to systemd.

>
> Can you please explain what is prompting to backport this change into
> v5.4 kernels?

For systems which don't have Internet access but do have valid RTC
time, if the RTC driver is a module, the RTC time will never be used
and instead UC20 would start from a fixed time which causes problems
in some projects.

Thanks,
Jesse

>
>
> --
> Regards,
>
> Dimitri.
>
> On Wed, Nov 24, 2021 at 11:01 AM Wen-chien Jesse Sung
> <jesse.sung at canonical.com> wrote:
> >
> > BugLink: https://launchpad.net/bugs/1952077
> >
> > == Impact ==
> > Although we have CONFIG_RTC_HCTOSYS enabled and set CONFIG_RTC_HCTOSYS_DEVICE
> > to "rtc0", system time would only be set if the RTC driver is built-in. This
> > may not be an issue on most amd64 systems because CONFIG_RTC_DRV_CMOS is y
> > for amd64, however it is an issue for other architectures since there may not
> > be a proper RTC driver enabled as built-in. Also making every RTC driver that
> > may be used as built-in is not a good option.
> >
> > == Fix ==
> > The commit is in mainline since v5.7:
> >
> > == Risk of Regression ==
> > It's a clean cherry-pick for v5.4 and looks straight forward so risk should be low.
> >
> >
> > Steve Muckle (1):
> >   rtc: class: support hctosys from modular RTC drivers
> >
> >  drivers/rtc/Kconfig   |  3 --
> >  drivers/rtc/Makefile  |  1 -
> >  drivers/rtc/class.c   | 61 ++++++++++++++++++++++++++++++++++++++
> >  drivers/rtc/hctosys.c | 69 -------------------------------------------
> >  4 files changed, 61 insertions(+), 73 deletions(-)
> >  delete mode 100644 drivers/rtc/hctosys.c
> >
> > --
> > 2.25.1
> >
> >
> > --
> > kernel-team mailing list
> > kernel-team at lists.ubuntu.com
> > https://lists.ubuntu.com/mailman/listinfo/kernel-team