cpufreq: Fix the race condition while updating the transition_task of policy
authorLiao Chang <liaochang1@huawei.com>
Tue, 29 Aug 2023 07:03:18 +0000 (07:03 +0000)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 29 Aug 2023 18:19:31 +0000 (20:19 +0200)
commit61bfbf7951ba561dcbdd5357702d3cbc2d447812
tree5885a51993f2268d2403143f307ed98b3dfd894c
parent1f464cb4a7febcc8b5ad325f3f9a83c8d4e4c199
cpufreq: Fix the race condition while updating the transition_task of policy

The field 'transition_task' of policy structure is used to track the
task which is performing the frequency transition. Using this field to
print a warning once detect a case where the same task is calling
_begin() again before completing the preivous frequency transition via
the _end().

However, there is a potential race condition in _end() and _begin() APIs
while updating the field 'transition_task' of policy, the scenario is
depicted below:

             Task A                            Task B

        /* 1st freq transition */
        Invoke _begin() {
                ...
                ...
        }
                                        /* 2nd freq transition */
                                        Invoke _begin() {
                                                ... //waiting for A to
                                                ... //clear
                                                ... //transition_ongoing
                                                ... //in _end() for
                                                ... //the 1st transition
                                                        |
        Change the frequency                            |
                                                        |
        Invoke _end() {                                 |
                ...                                     |
                ...                                     |
                transition_ongoing = false;             V
                                                transition_ongoing = true;
                                                transition_task = current;
                transition_task = NULL;
                ... //A overwrites the task
                ... //performing the transition
                ... //result in error warning.
        }

To fix this race condition, the transition_lock of policy structure is
now acquired before updating policy structure in _end() API. Which ensure
that only one task can update the 'transition_task' field at a time.

Link: https://lore.kernel.org/all/b3c61d8a-d52d-3136-fbf0-d1de9f1ba411@huawei.com/
Fixes: ca654dc3a93d ("cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end")
Signed-off-by: Liao Chang <liaochang1@huawei.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/cpufreq/cpufreq.c