Re: [PATCH] cpufreq: acpi-cpufreq: add NULL check for acpi_perf_data before freeing

On 8/10/2026 7:21 PM, Zhongqiu Han wrote:
> On 8/10/2026 2:10 PM, lirongqing wrote:
>> From: Li RongQing <[email protected]>
>>
>> If alloc_percpu() fails in acpi_cpufreq_early_init(), acpi_perf_data is
>> NULL.  Calling free_acpi_perf_data() will execute per_cpu_ptr(NULL, i),
>> which does not return NULL but an offset pointer.
> 
> Hi RongQing,
> 
> Could you please confirm if acpi_perf_data() can ever be executed when 
> alloc_percpu() fails?

Sorry, typo: acpi_perf_data() --> free_acpi_perf_data()

> 
> 
>> Dereferencing ->shared_cpu_map on this invalid pointer leads to a kernel
>> crash.
>>
>> Fix these issues by adding a NULL check at the start of
>> free_acpi_perf_data() and setting acpi_perf_data to NULL after freeing.
>>
>> Signed-off-by: Li RongQing <[email protected]>
>> ---
>>   drivers/cpufreq/acpi-cpufreq.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi- 
>> cpufreq.c
>> index 87e4923..cd2ca87 100644
>> --- a/drivers/cpufreq/acpi-cpufreq.c
>> +++ b/drivers/cpufreq/acpi-cpufreq.c
>> @@ -517,11 +517,14 @@ static void free_acpi_perf_data(void)
>>   {
>>       unsigned int i;
>> -    /* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
>> +    if (!acpi_perf_data)
>> +        return;
>> +
>>       for_each_possible_cpu(i)
>>           free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
>>                    ->shared_cpu_map);
>>       free_percpu(acpi_perf_data);
>> +    acpi_perf_data = NULL;
>>   }
>>   static int cpufreq_boost_down_prep(unsigned int cpu)
> 
> 


-- 
Thx and BRs,
Zhongqiu Han