Vadym,

what's the issue here? 'halt -f' should immediately halt the system without going through shutdown. Is that what you saw? Or something else?

Thanks,
Andrew

Exactly, it "should immediately halt" it.
Instead, it powers it off, like it was called with -p switch.
halt and power off is a big difference.

Testing 'halt -f' on all RHEL 5.5 guest configs 32/64-bit PV/HV it works for me, and so does issuing 'halt -f' from 64-bit dom0. Can you check to see if you still have the problem when using 5.5?

Andrew

yes, it's still the case, that's why updated the version number to 5.5

When halt -f is issued in dom0 domain, instead of halting the server, 
it powers it off.
As a workaround I always create a /sbin/halt.local, like this:

#!/bin/bash

# See if this is a powerfail situation.
if [ -f /etc/apcupsd/powerfail ]; then

echo
echo "APCUPSD will now power off the UPS"
echo

/etc/apcupsd/apccontrol killpower

# Workaround for bz# 475006
while :
do
  echo Halt
  sleep 10
done
fi

If server was powered off, instead of halted, 
it will not come online when power is restored.

And this only happens with xen kernels? bare-metal on the same hardware behaves as expected?

Correct

The UPS seems to be important as it doesn't reproduce here without.  Can you trigger it without one?

It is importnat for power recovery, but UPS presence is not important.

It seems the problem doesn't exist in 5.6 kernel

OK, as we haven't seen this issue reported broadly, and it seems to have gone away for the reporter as well with 5.6, then I believe it's safe to close this bug as current release.