Re: [PATCH] platform/x86: intel_pmc_core: fix out-of-bounds accesses on stack

[Andrey Ryabinin]

On 01/27/2017 06:42 PM, Andy Shevchenko wrote:
> On Thu, Jan 26, 2017 at 4:27 PM, Andrey Ryabinin
> <aryabinin@xxxxxxxxxxxxx> wrote:


>>  #if IS_ENABLED(CONFIG_DEBUG_FS)
>> @@ -234,12 +230,8 @@ static const struct file_operations pmc_core_ppfear_ops = {
>>  /* This function should return link status, 0 means ready */
>>  static int pmc_core_mtpmc_link_status(void)
>>  {
>> -       struct pmc_dev *pmcdev = &pmc;
>> -       u32 value;
>> -
>> -       value = pmc_core_reg_read(pmcdev, SPT_PMC_PM_STS_OFFSET);
>> -       return test_bit(SPT_PMC_MSG_FULL_STS_BIT,
>> -                       (unsigned long *)&value);
>> +       u32 value = pmc_core_reg_read(&pmc, SPT_PMC_PM_STS_OFFSET);
>> +       return value & (1U << SPT_PMC_MSG_FULL_STS_BIT);
>>  }
> 
> Thanks for the patch. IIRC I told (or may be forgot to tell) them
> during internal review about the nasty casting.
> 
> Btw, have you checked this will work in the same way, since test_bit()
> is atomic?

'value' is a local variable, atomicity is pointless here.

> And if it's okay, why not to use BIT() macro?
> 

It just a matter of taste. I find open-coded variant easier to read.