Current code looks like this:
WARN_ON(lock_policy_rwsem_write(cpu)); update_policy_cpu(policy, new_cpu); unlock_policy_rwsem_write(cpu);
{lock|unlock}_policy_rwsem_write(cpu) takes/releases policy->cpu's rwsem. Because cpu is changing with the call to update_policy_cpu(), the unlock_policy_rwsem_write() will release the incorrect lock.
The right solution would be to release the same lock as was taken earlier. Also update_policy_cpu() was also called from cpufreq_add_dev() without any locks and so its better if we move this locking to inside update_policy_cpu().
Reported-and-Tested-by: Jon Medhursttixy@linaro.org Signed-off-by: Viresh Kumar viresh.kumar@linaro.org --- Hi Rafael,
Only one patch is sent now as other one is unchanged.
drivers/cpufreq/cpufreq.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index 43c24aa..1479522 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -952,9 +952,20 @@ static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu) if (cpu == policy->cpu) return;
+ /* + * Take direct locks as lock_policy_rwsem_write wouldn't work here. + * Also lock for last cpu is enough here as contention will happen only + * after policy->cpu is changed and after it is changed, other threads + * will try to acquire lock for new cpu. And policy is already updated + * by then. + */ + down_write(&per_cpu(cpu_policy_rwsem, policy->cpu)); + policy->last_cpu = policy->cpu; policy->cpu = cpu;
+ up_write(&per_cpu(cpu_policy_rwsem, policy->last_cpu)); + #ifdef CONFIG_CPU_FREQ_TABLE cpufreq_frequency_table_update_policy_cpu(policy); #endif @@ -1203,9 +1214,7 @@ static int __cpufreq_remove_dev_prepare(struct device *dev,
new_cpu = cpufreq_nominate_new_policy_cpu(policy, cpu, frozen); if (new_cpu >= 0) { - WARN_ON(lock_policy_rwsem_write(cpu)); update_policy_cpu(policy, new_cpu); - unlock_policy_rwsem_write(cpu);
if (!frozen) { pr_debug("%s: policy Kobject moved to cpu: %d "