CPU hotplug error handling fixes

This series of patches fixes the error handling for cpu hotplug. The problem is revealed by CPU hotplug/unplug test with fault-injection.

The patch 1-3 are sysfs or driver core related error handling fixes. These are not directly related to cpu hotplug. But these are needed to pass the stress test.

The patch 4 changes the behavior when one of the callbacks in notifier chain returns NOTIFY_BAD with CPU_UP_PREPARE event. This change makes cpu hotplug error handling simple.

The patch 5 simplifies the cpu hotplug event handling in topology.c by the patch 4.

The patch 6-10 are error handling fixes in cpu hotplug event callbacks. These fixes also depend on the change by the patch 4.

Here is the test script I have confirmed with these patches. I guess we still have the similar bugs that I could not test due to no hardware. So it may be worth someone trying this script.

----------[ cut here ]----------

#!/bin/bash

FAILTYPE=failslab
CPU=1 CPU_ONLINE=/sys/devices/system/cpu/cpu${CPU}/online

faulty_system()

{
        bash -c "echo 1 > /proc/self/make-it-fail && exec $*"
}

[ "$UID" == 0 ] || exit 1
[ -n "$FAILTYPE" -a -f /debug/$FAILTYPE/probability ] || exit 1 [ -f $CPU_ONLINE ] || exit 1

echo N > /debug/$FAILTYPE/ignore-gfp-wait
echo Y > /debug/$FAILTYPE/task-filter
echo 1 > /debug/$FAILTYPE/probability
echo -1 > /debug/$FAILTYPE/times

while true
do
faulty_system "echo 0 > $CPU_ONLINE"
faulty_system "echo 1 > $CPU_ONLINE"
done