2022-05-09 05:09:02

by Schspa Shi

Subject: [PATCH] cpufreq: fix double unlock when cpufreq online

The patch f346e96267cd: ("cpufreq: Fix possible race in cpufreq online
error path") expand the critical region. But policy->rwsem is not held when
calling cpufreq_driver->online and cpufreq_driver->init calls, which lead to bad
unlock.

And it's well to hold this lock when calling cpufreq_driver->online, which
provide more protects without bad influence.

Fixes: f346e96267cd: ("cpufreq: Fix possible race in cpufreq online error path")
Link: https://lore.kernel.org/all/YnKZCGaig+EXSowf@kili/

Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Schspa Shi <[email protected]>
---
drivers/cpufreq/cpufreq.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 0d58b0f8f3af..43dfaa8124e2 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -1337,12 +1337,12 @@ static int cpufreq_online(unsigned int cpu)
down_write(&policy->rwsem);
policy->cpu = cpu;
policy->governor = NULL;
- up_write(&policy->rwsem);
} else {
new_policy = true;
policy = cpufreq_policy_alloc(cpu);
if (!policy)
return -ENOMEM;
+ down_write(&policy->rwsem);
}

if (!new_policy && cpufreq_driver->online) {
@@ -1382,7 +1382,6 @@ static int cpufreq_online(unsigned int cpu)
cpumask_copy(policy->related_cpus, policy->cpus);
}

- down_write(&policy->rwsem);
/*
* affected cpus must always be the one, which are online. We aren't
* managing offline cpus here.
@@ -1542,9 +1541,9 @@ static int cpufreq_online(unsigned int cpu)
cpufreq_driver->exit(policy);

cpumask_clear(policy->cpus);
- up_write(&policy->rwsem);

out_free_policy:
+ up_write(&policy->rwsem);
cpufreq_policy_free(policy);
return ret;
}
--
2.24.3 (Apple Git-128)

2022-05-09 10:09:10

by Viresh Kumar

Subject: Re: [PATCH] cpufreq: fix double unlock when cpufreq online

On 07-05-22, 01:00, Schspa Shi wrote:
> The patch f346e96267cd: ("cpufreq: Fix possible race in cpufreq online
> error path") expand the critical region. But policy->rwsem is not held when
> calling cpufreq_driver->online and cpufreq_driver->init calls, which lead to bad
> unlock.
>
> And it's well to hold this lock when calling cpufreq_driver->online, which
> provide more protects without bad influence.
>
> Fixes: f346e96267cd: ("cpufreq: Fix possible race in cpufreq online error path")
> Link: https://lore.kernel.org/all/YnKZCGaig+EXSowf@kili/
>
> Reported-by: Dan Carpenter <[email protected]>
> Signed-off-by: Schspa Shi <[email protected]>
> ---
> drivers/cpufreq/cpufreq.c | 5 ++---
> 1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 0d58b0f8f3af..43dfaa8124e2 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -1337,12 +1337,12 @@ static int cpufreq_online(unsigned int cpu)
> down_write(&policy->rwsem);
> policy->cpu = cpu;
> policy->governor = NULL;
> - up_write(&policy->rwsem);
> } else {
> new_policy = true;
> policy = cpufreq_policy_alloc(cpu);
> if (!policy)
> return -ENOMEM;
> + down_write(&policy->rwsem);
> }
>
> if (!new_policy && cpufreq_driver->online) {
> @@ -1382,7 +1382,6 @@ static int cpufreq_online(unsigned int cpu)
> cpumask_copy(policy->related_cpus, policy->cpus);
> }
>
> - down_write(&policy->rwsem);
> /*
> * affected cpus must always be the one, which are online. We aren't
> * managing offline cpus here.
> @@ -1542,9 +1541,9 @@ static int cpufreq_online(unsigned int cpu)
> cpufreq_driver->exit(policy);
>
> cpumask_clear(policy->cpus);
> - up_write(&policy->rwsem);
>
> out_free_policy:
> + up_write(&policy->rwsem);
> cpufreq_policy_free(policy);
> return ret;
> }

Since this is a tricky piece of code, I suggest sending a fresh patch
to fix the original issue over the revert I have sent. I am not yet
sure both patches combined will fix it correctly.

--
viresh