PM / Sleep: Fix race between CPU hotplug and freezer
authorSrivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Wed, 2 Nov 2011 23:59:25 +0000 (00:59 +0100)
committerRafael J. Wysocki <rjw@sisk.pl>
Fri, 4 Nov 2011 21:28:09 +0000 (22:28 +0100)
commit79cfbdfa87e84992d509e6c1648a18e1d7e68c20
treecc82be9bc890aa0b2b635ceff06db2ed9056460d
parentddf6ce45a7b1193f3cf20ad234f35af3b998b8f8
PM / Sleep: Fix race between CPU hotplug and freezer

The CPU hotplug notifications sent out by the _cpu_up() and _cpu_down()
functions depend on the value of the 'tasks_frozen' argument passed to them
(which indicates whether tasks have been frozen or not).
(Examples for such CPU hotplug notifications: CPU_ONLINE, CPU_ONLINE_FROZEN,
CPU_DEAD, CPU_DEAD_FROZEN).

Thus, it is essential that while the callbacks for those notifications are
running, the state of the system with respect to the tasks being frozen or
not remains unchanged, *throughout that duration*. Hence there is a need for
synchronizing the CPU hotplug code with the freezer subsystem.

Since the freezer is involved only in the Suspend/Hibernate call paths, this
patch hooks the CPU hotplug code to the suspend/hibernate notifiers
PM_[SUSPEND|HIBERNATE]_PREPARE and PM_POST_[SUSPEND|HIBERNATE] to prevent
the race between CPU hotplug and freezer, thus ensuring that CPU hotplug
notifications will always be run with the state of the system really being
what the notifications indicate, _throughout_ their execution time.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
kernel/cpu.c