PM: Do not hold dpm_list_mtx while disabling/enabling nonboot CPUs
authorRafael J. Wysocki <rjw@sisk.pl>
Sun, 24 May 2009 19:15:07 +0000 (21:15 +0200)
committerRafael J. Wysocki <rjw@sisk.pl>
Sun, 24 May 2009 19:15:07 +0000 (21:15 +0200)
commit32bdfac5462d777f35b00838893c4f87baf23efe
tree92e4ef3af7b68007e8004eaca978865a29e543b0
parent59a3759d0fe8d969888c741bb33f4946e4d3750d
PM: Do not hold dpm_list_mtx while disabling/enabling nonboot CPUs

We shouldn't hold dpm_list_mtx while executing
[disable|enable]_nonboot_cpus(), because theoretically this may lead
to a deadlock as shown by the following example (provided by Johannes
Berg):

CPU 3       CPU 2                     CPU 1
                                      suspend/hibernate
            something:
            rtnl_lock()               device_pm_lock()
                                       -> mutex_lock(&dpm_list_mtx)

            mutex_lock(&dpm_list_mtx)

linkwatch_work
 -> rtnl_lock()
                                      disable_nonboot_cpus()
                                       -> flush CPU 3 workqueue

Fortunately, device drivers are supposed to stop any activities that
might lead to the registration of new device objects way before
disable_nonboot_cpus() is called, so it shouldn't be necessary to
hold dpm_list_mtx over the entire late part of device suspend and
early part of device resume.

Thus, during the late suspend and the early resume of devices acquire
dpm_list_mtx only when dpm_list is going to be traversed and release
it right after that.

This patch is reported to fix the regressions tracked as
http://bugzilla.kernel.org/show_bug.cgi?id=13245.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Miles Lane <miles.lane@gmail.com>
Tested-by: Ming Lei <tom.leiming@gmail.com>
drivers/base/power/main.c
kernel/kexec.c
kernel/power/disk.c
kernel/power/main.c