After upgrade to Lenny I experience reproducible complete lockups when X Window is running.
Narrowing this problem down made it clear that the culprit is
chrony with RTC adjustments enabled ("rtcfile" directive
uncommented in the configuration file).
The system freezes rather strangely -- PS/2 devices (mouse and
keyboard) lock up for sure while sound may still work and video
rendering also sometimes isn't affected). Without X Windows running
the system doesn't freeze (at lest not in ~24 hours of run).
Anyway it's surely chrony, since with it not running or running
with "rtcfile" commented out the system works OK.
The problem is that:
* It worked OK on Etch with the very same hardware.
* The trickery with RTC kernel modules mentioned in the first few
comments on this bug no longer works because RTC support seems to
be compiled in now -- at least `lsmod|grep rtc` yields nothing
with chronyd running. `modinfo rtc` and `modinfo genrtc`
also show nothing.
Also in a couple of seconds after chronyd is started with RTC handling
enabled (the actual timeout varies) the kernel reports that it lost
several RTC interrupts, like this:
[ 443.791483] rtc: lost 3 interrupts
There may be several messages in a row, and a number of interrupts
also varies.
Also there may be no such messages reported before a freeze, but
they're never reported if chronyd isn't running or has RTC handling
disabled.
This bug is known to happen on "HPET" systems and I do have
this message reported by the kernel:
[ 0.272693] pci 0000:00:1f.0: Force enabled HPET at 0xfed00000
I googled this discussion [1] which says this bug should go away if
CONFIG_HPET_EMULATE_RTC kernel configuration option is turned on
but it seems not to work in my case:
$ grep HPET /boot/config-2.6.26-1-686
CONFIG_HPET_TIMER=y
CONFIG_HPET_EMULATE_RTC=y
CONFIG_HPET=y
CONFIG_HPET_MMAP=y
The discussion reported above mentions HPET bug in 2.6.26 [2]
which is the Lenny kernel's version.
I don't know whether it is integrated in the Lenny kernel or
whether it may help at all though since I have absolutely no
expertise in RTC subsystem.
In the meantime my proposition is to comment out the "rtcfile" in
the config file installed by chrony and make a more bold warning
about the possible complications which may occur after enabling it.
I have attached the dmesg and lshw outputs on my system as well as
a strace report of chronyd after it gets several NTP sources
available and seems to start normal operation.
[1]: https://qa.mandriva.com/show_bug.cgi?id=42106#c2
[2]: http://bugzilla.kernel.org/show_bug.cgi?id=11112
dmesg.gz
Description: dmesg output
chrony.2869.gz
Description: strace'ing of a running chronyd with RTC handling enabled
lshw.gz
Description: output of lshw
