Re: [PATCH] acpi-cpufreq: Add a miss ifdef CONFIG_X86_ACPI_CPUFREQ_CPB

[Pan Xinhui <xinhuix.pan@intel.com>]

hi, Rafael,
	let me do more explanation :)

On 2015年07月14日 10:09, Pan Xinhui wrote:
> hi, Rafael,
> 	thanks for you reply :)
> On 2015年07月14日 07:26, Rafael J. Wysocki wrote:
>> On Monday, July 13, 2015 02:33:08 PM Pan Xinhui wrote:
>>> hi, Rafeal
>>> 	thanks for your reply. :)
>>>
>>> On 2015年07月11日 04:44, Rafael J. Wysocki wrote:
>>>> Hi,
>>>>
>>>> On Fri, Jul 10, 2015 at 7:50 AM, Pan Xinhui <xinhuix.pan@intel.com> wrote:
>>>>>
>>>>> If CONFIG_X86_ACPI_CPUFREQ_CPB has not been defined, the placeholder for
>>>>> cpb is not needed. Add ifdef around it.
>>>>>
>>>>> Signed-off-by: Pan Xinhui <xinhuix.pan@intel.com>
>>>>> ---
>>>>>  drivers/cpufreq/acpi-cpufreq.c | 2 ++
>>>>>  1 file changed, 2 insertions(+)
>>>>>
>>>>> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
>>>>> index e7fcaa6..314a19e 100644
>>>>> --- a/drivers/cpufreq/acpi-cpufreq.c
>>>>> +++ b/drivers/cpufreq/acpi-cpufreq.c
>>>>> @@ -884,7 +884,9 @@ static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
>>>>>  static struct freq_attr *acpi_cpufreq_attr[] = {
>>>>>         &cpufreq_freq_attr_scaling_available_freqs,
>>>>>         &freqdomain_cpus,
>>>>> +#ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
>>>>>         NULL,   /* this is a placeholder for cpb, do not remove */
>>>>> +#endif
>>>>
>>>> Adding the ifdef here doesn't change anything, because the next NULL
>>>> will play the role of the one you've just #ifdefed and the structure
>>>> will be filled with zeros from that point on anyway.
>>>>
>>> Yes, adding ifdef here does not change any binary codes. But I want to make the codes more readable. :)
>>> Patch author has noticed two *NULL* here would confuse people, especially who first read this acpi-cpufreq.c file
>>> From code style point, it would be better to have #ifdef around it. 
>>
>> Not really.
>>
>> Why don't you simply drop *both* NULLs?
>>
> Just like string end with *NULL* :)
> 
> 1021 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy,
> 1022                      struct device *dev)
> 1023 {
> 1024     struct freq_attr **drv_attr;
> 1025     int ret = 0;
> 1026 
> 1027     /* set up files for this cpu device */
> 1028     drv_attr = cpufreq_driver->attr;
> 1029     while (drv_attr && *drv_attr) {
> 1030         ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
> 1031         if (ret)
> 1032             return ret;
> 1033         drv_attr++;
> 1034     }
> If struct freq_attr *acpi_cpufreq_attr[] did not end with NULL, line 1033 will access invalid data area.
> If *drv_attr(the data after struct freq_attr * array[]) happened to be not NULL. panic may hit in sysfs_create_file :(
> So at least one *NULL* must be in the end of freq_attr *array[].
> 
> Actually in acpi-cpufreq.c, in acpi_cpufreq_init function.
>  957         struct freq_attr **iter;
>  958 
>  959         pr_debug("adding sysfs entry for cpb\n");
>  960 
>  961         for (iter = acpi_cpufreq_attr; *iter != NULL; iter++)
>  962             ;
>  963 
>  964         /* make sure there is a terminator behind it */
>  965         if (iter[1] == NULL)
>  966             *iter = &cpb;
>  967     }
> line965, check of iter[1] is not needed. Maybe the patch author was afraid of an unexpected remove of first *NULL*.
> It might be a better solution to add ifdef CONFIG_X86_ACPI_CPUFREQ_CPB around that *NULL*, and remove this !iter[1] check.
> 
According to line 961, the for loop, patch author assume the first *NULL* may be *at any index* of struct freq_attr *acpi_cpufreq_attr[].

But according to line 965, and the comment also says *a terminator behind it*. patch author assume the first *NULL* is *just right before* the second *NULL* which acts as a terminator.
The logic is incorrect. :(
Why not just
+		/* there must be two NULL in the end of acpi_cpufreq_attr[] */
+               struct freq_attr **iter = &acpi_cpufreq_attr[sizeof(acpi_cpufreq_attr) - 2];
+               if (!WARN_ON(iter[0] || iter[1]))

I don't like that codes assume the index of first *NULL*, so maybe removing line 965, the check if iter[1] == NULL, is more correct and beautiful :)


thanks
xinhui

> thanks
> xinhui
> 
>>
>>>
>>>> You'd need to #ifdef it in the struct freq_attr definition, but I'm
>>>> not sure it's worth the effort.
>>>>
>>>
>>> struct freq_attr *cpb* is defined in #ifdef section. :)
>>
>> Ah, OK.
>>
>>