PM-runtime: Fix __pm_runtime_set_status() race with runtime resume
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 12 Feb 2019 12:04:12 +0000 (13:04 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 13 Feb 2019 08:09:51 +0000 (09:09 +0100)
commitc1567f813a9992a64f8d0f6cfb912c3922812c35
tree9401ab62aa84312f6c9f66674b8de4b4dab4b373
parentdaaef255dc96834aaaad627d3271504cba3ac2dc
PM-runtime: Fix __pm_runtime_set_status() race with runtime resume

Commit 4080ab083000 ("PM-runtime: Take suppliers into account in
__pm_runtime_set_status()") introduced a race condition that may
trigger if __pm_runtime_set_status() is used incorrectly (that is,
if it is called when PM-runtime is enabled for the target device
and working).

In that case, if the original PM-runtime status of the device is
RPM_SUSPENDED, a runtime resume of the device may occur after
__pm_runtime_set_status() has dropped its power.lock spinlock
and before deactivating its suppliers, so the suppliers may be
deactivated while the device is PM-runtime-active which may lead
to functional issues.

To avoid that, modify __pm_runtime_set_status() to check whether
or not PM-runtime is enabled for the device before activating its
suppliers (if the new status is RPM_ACTIVE) and either return an
error if that's the case or increment the device's disable_depth
counter to prevent PM-runtime from being enabled for it while
the remaining part of the function is running (disable_depth is
then decremented on the way out).

Fixes: 4080ab083000 ("PM-runtime: Take suppliers into account in __pm_runtime_set_status()")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/power/runtime.c