W dniu 01.03.2021 o 17:19, Lennart Poettering pisze: > On Mo, 01.03.21 17:09, Michał Zegan (webczat_200@xxxxxxxxxxxxxx) wrote: > >>>> There are problems with log timestamps when you do that, and it is >>>> probably why it was not done. >>>> I am wondering if the only correct way isn't to do it in initramfs (if >>>> it's systemd) before starting the journald, so that first saved logs >>>> have correct timestamps? >>> >>> The earlier you have a "correct" clock the better. >>> >>> But I mean, journald won't lie to you: the log messages are associated >>> with the timestamps that were accurate at the moment they are >>> generated. But of course, if your clock sucks then they'll look >>> differently than you might expect. >> However the initial log entries in journald are taken from dmesg and >> dmesg timestamps are relative, so setting the clock before journald >> first starts should make journal times correct. >> Or I am wrong? > > kmsg comes with monotonic timestamps only. journald stores that away > but generally uses its own acquired timestamps, since it needs to > guarantee monotonicity and things. > > But do note that journald typically runs in the initrd already. Yes, my idea was to wait for time *in initramfs* before journald. Like force the rtc driver to land in initramfs, then just load it, set time if kernel doesn't, and then start journald. > > Lennart > > -- > Lennart Poettering, Berlin >
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
_______________________________________________ systemd-devel mailing list systemd-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/systemd-devel
