[systemd-devel] CAD often useless

[Lennart Poettering]

On Wed, 15.04.15 18:29, Felix Miata (mrmazda at earthlink.net) wrote:

> Felix Miata composed on 2015-04-15 14:48 (UTC-0400):
> 
> > Zbigniew Jędrzejewski-Szmek composed on 2015-04-15 18:11 (UTC):
> > 
> >> On Wed, Apr 15, 2015 at 13:31:38 -0400, Felix Miata wrote:
> > 
> >>> This isn't the first time or the only system. This particular one is an old
> >>> Athlon booted to F22 just updated last night. In order to try some follow-up
> >>> on a bug, I booted with drm.debug=14 log_buf_len=16M on cmdline. Somehow
> >>> after exiting IceWM back into KDM, the system quit responding. After some
> >>> waiting, I was switched to a tty full of systemd-journal failed to write
> >>> readonly filesystem messages. How the system got / into a readonly state I
> >>> have no idea, but my complaint is $SUBJECT, the time it takes, or the
> >>> complete failure, trying to reboot when there is a problem, often seeing
> >>> failed to store sound card state or failing boot *start* jobs. Why has
> >>> rebooting to get out of trouble gotten to be so nearly impossible?
> 
> >> Did you try to press CAD more than 7 times within 2s? This should
> >> result in immedate reboot.
> 
> > My fingers and keyboards are old. I just hold CAD down until reboot proceeds,
> > or not. All that have repeat and delay configurable in BIOS are set to 20 cps
> > & 250 delay. KDE I always set to same.
> 
> Needing again to CAD on yet another machine (kt88b this time, kt400 earlier),
> again encountering / going into RO state, I noticed this time the message
> "Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting
> immediately". I remember having seen this on previous occasions to need CAD,
> but it's still on the screen now, several minutes later, with the NUM key's
> LED remaining responsive. IOW, it doesn't do either what it says, or what I
> wanted. This is on F22 with PAE kernel 4.0.0-1 and PATA storage only except
> for empty floppy drive.

When you see this message, then systemd will not bother with shuttding
down daemons cleanly anymore. However, it will still unmount all file
systems and sync things do disk. It will also pass control back to the
initrd, which might disassemble LVM if that is used.

If things hang in that logic then this is either an issue with
LVM/initrd, or with your driver/hw when we try to sync the disks.

Given that your fs was remounted read-only (which the kernel does
automatically if it notices something seriously wrong with the disk),
my guess is that the disk syncing trips you up.

Lennart

-- 
Lennart Poettering, Red Hat