Re: [PATCH] drivers/acpi/processor_driver.c: add missing kfree

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 03/14/2012 06:02 PM, Julia Lawall wrote:

> 
> 
> On Wed, 14 Mar 2012, Srivatsa S. Bhat wrote:
> 
>> On 03/10/2012 10:35 PM, Julia Lawall wrote:
>>
>>> From: Julia Lawall <julia@diku.dk>
>>>
>>> The function acpi_processor_add is stored in the ops.add field of a
>>> acpi_driver structure.  This function is then called in
>>> acpi_bus_driver_init.  On failure, this function clears the field
>>> device->driver_data, but does not free its contents.  Thus the free
>>> has to
>>> be done by the add function.  In acpi_processor_add, the corresponding
>>> value is pr.  This value is currently freed on failure before storing
>>> it in
>>> device->driver_data, but not after.  This free is added in the error
>>> handling code at the end of the function.  The static global variable
>>
>>
>> "static global variable"?? never heard that one before ;-)
>> Maybe you meant "per_cpu variable processors"..
>>
>>> processors is also cleared so that it does not refer to a dangling
>>> pointer.
>>> processor_device_array is cleared as well.
>>>
>>> Signed-off-by: Julia Lawall <julia@diku.dk>
>>>
>>> ---
>>> This is only compile tested.  In particular, I don't know if it is
>>> correct
>>> to add per_cpu(processor_device_array, pr->id) = NULL;.
>>
>>
>> No, you shouldn't set it to NULL. processor_device_array was added to
>> check
>> for buggy BIOSes and return gracefully. Check commit cd8e2b48d (and
>> also the
>> bugzilla link in that commit).
>>
>> To have a robust check for buggy BIOSes, we must let it be as it is,
>> even if
>> it is a stale pointer. That way we can still catch subsequent calls to
>> this
>> function with the same acpi id (because of a buggy BIOS) and take
>> appropriate
>> actions.
>>
>> Other than that, the patch looks good to me.
> 
> Thanks for the feedback.  Just to be clear, I should keep
> 
> +err_clear_processors:
> +       per_cpu(processors, pr->id) = NULL;
> 
> and just drop:
> 
> +       per_cpu(processor_device_array, pr->id) = NULL;
> 


Yep, that's right. But in fact, you can even do better than that...
That is, drop the above line and put a comment that explains why we
shouldn't set per_cpu(processor_device_array..) to NULL. That way,
in future people will know that setting it to NULL was left out on
purpose.

And please adjust the commit message too, as I pointed in my previous
mail :-)

Regards,
Srivatsa S. Bhat

>>>
>>>  drivers/acpi/processor_driver.c |   12 ++++++++----
>>>  1 file changed, 8 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/acpi/processor_driver.c
>>> b/drivers/acpi/processor_driver.c
>>> index 2801b41..9bb0017 100644
>>> --- a/drivers/acpi/processor_driver.c
>>> +++ b/drivers/acpi/processor_driver.c
>>> @@ -536,8 +536,8 @@ static int __cpuinit acpi_processor_add(struct
>>> acpi_device *device)
>>>          return -ENOMEM;
>>>
>>>      if (!zalloc_cpumask_var(&pr->throttling.shared_cpu_map,
>>> GFP_KERNEL)) {
>>> -        kfree(pr);
>>> -        return -ENOMEM;
>>> +        result = -ENOMEM;
>>> +        goto err_free_pr;
>>>      }
>>>
>>>      pr->handle = device->handle;
>>> @@ -577,7 +577,7 @@ static int __cpuinit acpi_processor_add(struct
>>> acpi_device *device)
>>>      dev = get_cpu_device(pr->id);
>>>      if (sysfs_create_link(&device->dev.kobj, &dev->kobj, "sysdev")) {
>>>          result = -EFAULT;
>>> -        goto err_free_cpumask;
>>> +        goto err_clear_processors;
>>>      }
>>>
>>>      /*
>>> @@ -595,9 +595,13 @@ static int __cpuinit acpi_processor_add(struct
>>> acpi_device *device)
>>>
>>>  err_remove_sysfs:
>>>      sysfs_remove_link(&device->dev.kobj, "sysdev");
>>> +err_clear_processors:
>>> +    per_cpu(processors, pr->id) = NULL;
>>> +    per_cpu(processor_device_array, pr->id) = NULL;
>>>  err_free_cpumask:
>>>      free_cpumask_var(pr->throttling.shared_cpu_map);
>>> -
>>> +err_free_pr:
>>> +    kfree(pr);
>>>      return result;
>>>  }
>>>
>>>
>>