在 2025/9/4 21:17, Rafael J. Wysocki 写道:
> On Thu, Sep 4, 2025 at 12:38 PM Zihuan Zhang <zhangzihuan@kylinos.cn> wrote:
>>
>> 在 2025/9/3 21:45, Rafael J. Wysocki 写道:
>>> On Wed, Sep 3, 2025 at 3:18 PM Zihuan Zhang <zhangzihuan@kylinos.cn> wrote:
>>>> Replace the manual cpufreq_cpu_put() with __free(put_cpufreq_policy)
>>>> annotation for policy references. This reduces the risk of reference
>>>> counting mistakes and aligns the code with the latest kernel style.
>>>>
>>>> No functional change intended.
>>>>
>>>> Signed-off-by: Zihuan Zhang <zhangzihuan@kylinos.cn>
>>>> ---
>>>>    drivers/powercap/dtpm_cpu.c | 30 +++++++++++-------------------
>>>>    1 file changed, 11 insertions(+), 19 deletions(-)
>>>>
>>>> diff --git a/drivers/powercap/dtpm_cpu.c b/drivers/powercap/dtpm_cpu.c
>>>> index 99390ec1481f..f76594185fa2 100644
>>>> --- a/drivers/powercap/dtpm_cpu.c
>>>> +++ b/drivers/powercap/dtpm_cpu.c
>>>> @@ -144,19 +144,17 @@ static int update_pd_power_uw(struct dtpm *dtpm)
>>>>    static void pd_release(struct dtpm *dtpm)
>>>>    {
>>>>           struct dtpm_cpu *dtpm_cpu = to_dtpm_cpu(dtpm);
>>>> -       struct cpufreq_policy *policy;
>>>>
>>>>           if (freq_qos_request_active(&dtpm_cpu->qos_req))
>>>>                   freq_qos_remove_request(&dtpm_cpu->qos_req);
>>>>
>>>> -       policy = cpufreq_cpu_get(dtpm_cpu->cpu);
>>>> -       if (policy) {
>>>> +       struct cpufreq_policy *policy __free(put_cpufreq_policy) =
>>>> +               cpufreq_cpu_get(dtpm_cpu->cpu);
>>>> +
>>>> +       if (policy)
>>>>                   for_each_cpu(dtpm_cpu->cpu, policy->related_cpus)
>>>>                           per_cpu(dtpm_per_cpu, dtpm_cpu->cpu) = NULL;
>>>>
>>>> -               cpufreq_cpu_put(policy);
>>>> -       }
>>>> -
>>>>           kfree(dtpm_cpu);
>>>>    }
>>>>
>>>> @@ -192,7 +190,6 @@ static int cpuhp_dtpm_cpu_online(unsigned int cpu)
>>>>    static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
>>>>    {
>>>>           struct dtpm_cpu *dtpm_cpu;
>>>> -       struct cpufreq_policy *policy;
>>>>           struct em_perf_state *table;
>>>>           struct em_perf_domain *pd;
>>>>           char name[CPUFREQ_NAME_LEN];
>>>> @@ -202,21 +199,19 @@ static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
>>>>           if (dtpm_cpu)
>>>>                   return 0;
>>>>
>>>> -       policy = cpufreq_cpu_get(cpu);
>>>> +       struct cpufreq_policy *policy __free(put_cpufreq_policy) =
>>>> +               cpufreq_cpu_get(cpu);
>>>> +
>>>>           if (!policy)
>>>>                   return 0;
>>>>
>>>>           pd = em_cpu_get(cpu);
>>>> -       if (!pd || em_is_artificial(pd)) {
>>>> -               ret = -EINVAL;
>>>> -               goto release_policy;
>>>> -       }
>>>> +       if (!pd || em_is_artificial(pd))
>>>> +               return -EINVAL;
>>>>
>>>>           dtpm_cpu = kzalloc(sizeof(*dtpm_cpu), GFP_KERNEL);
>>>> -       if (!dtpm_cpu) {
>>>> -               ret = -ENOMEM;
>>>> -               goto release_policy;
>>>> -       }
>>>> +       if (!dtpm_cpu)
>>>> +               return -ENOMEM;
>>>>
>>>>           dtpm_init(&dtpm_cpu->dtpm, &dtpm_ops);
>>>>           dtpm_cpu->cpu = cpu;
>>>> @@ -239,7 +234,6 @@ static int __dtpm_cpu_setup(int cpu, struct dtpm *parent)
>>>>           if (ret < 0)
>>>>                   goto out_dtpm_unregister;
>>> So this change kind of goes against another recommendation given in cleanup.h:
>>>
>>>    * Lastly, given that the benefit of cleanup helpers is removal of
>>>    * "goto", and that the "goto" statement can jump between scopes, the
>>>    * expectation is that usage of "goto" and cleanup helpers is never
>>>    * mixed in the same function. I.e. for a given routine, convert all
>>>    * resources that need a "goto" cleanup to scope-based cleanup, or
>>>    * convert none of them.
>>
>> Should I replace all the memory allocation cleanups here with `__free`?
>> That would allow us to drop all the `goto`s, but since this function has
>> quite a few of them, I’m concerned it might introduce new issues. What’s
>> your recommendation?
> Frankly, don't use __free() in this code at all, at least for the time being.
>
> There is a problem with dropping the reference to policy at the end of
> __dtpm_cpu_setup() because that policy may be subsequently indirectly
> used in set_pd_power_limit() which calls
> freq_qos_update_request(&dtpm_cpu->qos_req, freq) and
> dtpm_cpu->qos_req->qos is policy->constraints, so using it will cause
> policy->constraints to be dereferenced in freq_qos_apply() which will
> crash and burn if the policy goes away in the meantime.  So AFAICS
> __dtpm_cpu_setup() shouldn't call cpufreq_cpu_put() at all and the
> policy should be released in pd_release() without acquiring a new
> reference to it.
>

Thanks a lot for the detailed explanation!