Re: [PATCH] hwmon: (smsc47m1) fix outside array bounds warnings

[Guenter Roeck <linux@roeck-us.net>]

On 5/27/19 8:09 PM, Masahiro Yamada wrote:
> On Tue, May 28, 2019 at 10:25 AM Guenter Roeck <linux@roeck-us.net> wrote:
>>
>> On 5/22/19 8:08 AM, Jean Delvare wrote:
>>> Hi Masahiro,
>>>
>>> On Tue, 21 May 2019 13:44:56 +0900, Masahiro Yamada wrote:
>>>> Kbuild test robot reports outside array bounds warnings:
>>>>
>>>>     CC [M]  drivers/hwmon/smsc47m1.o
>>>> drivers/hwmon/smsc47m1.c: In function 'fan_div_store':
>>>> drivers/hwmon/smsc47m1.c:370:49: warning: array subscript [0, 2] is outside array bounds of 'u8[3]' {aka 'unsigned char[3]'} [-Warray-bounds]
>>>>     tmp = 192 - (old_div * (192 - data->fan_preload[nr])
>>>>                                   ~~~~~~~~~~~~~~~~~^~~~
>>>> drivers/hwmon/smsc47m1.c:372:19: warning: array subscript [0, 2] is outside array bounds of 'u8[3]' {aka 'unsigned char[3]'} [-Warray-bounds]
>>>>     data->fan_preload[nr] = clamp_val(tmp, 0, 191);
>>>>     ~~~~~~~~~~~~~~~~~^~~~
>>>> drivers/hwmon/smsc47m1.c:373:53: warning: array subscript [0, 2] is outside array bounds of 'const u8[3]' {aka 'const unsigned char[3]'} [-Warray-bounds]
>>>>     smsc47m1_write_value(data, SMSC47M1_REG_FAN_PRELOAD[nr],
>>>>                                ~~~~~~~~~~~~~~~~~~~~~~~~^~~~
>>>
>>> These messages are pretty confusing. Subscript [0, 2] would refer to a
>>> bi-dimensional array, while these are 1-dimension arrays. If [0, 2]
>>> means something else, I still don't get it, because both indexes 0 and
>>> 2 are perfectly within bounds of a 3-element array. So what do these
>>> messages mean exactly? Looks like a bogus checker to me.
>>>
>>>> The index field in the SENSOR_DEVICE_ATTR_R* defines is 0, 1, or 2.
>>>> However, the compiler never knows the fact that the default in the
>>>> switch statement is unreachable.
>>>>
>>>> Reported-by: kbuild test robot <lkp@intel.com>
>>>> Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
>>>> ---
>>>>
>>>>    drivers/hwmon/smsc47m1.c | 4 ++++
>>>>    1 file changed, 4 insertions(+)
>>>>
>>>> diff --git a/drivers/hwmon/smsc47m1.c b/drivers/hwmon/smsc47m1.c
>>>> index 5f92eab24c62..e00102e05666 100644
>>>> --- a/drivers/hwmon/smsc47m1.c
>>>> +++ b/drivers/hwmon/smsc47m1.c
>>>> @@ -364,6 +364,10 @@ static ssize_t fan_div_store(struct device *dev,
>>>>               tmp |= data->fan_div[2] << 4;
>>>>               smsc47m1_write_value(data, SMSC47M2_REG_FANDIV3, tmp);
>>>>               break;
>>>> +    default:
>>>> +            WARN_ON(1);
>>>> +            mutex_unlock(&data->update_lock);
>>>> +            return -EINVAL;
>>>>       }
>>>
>>> So basically the code is fine, the checker (which checker, BTW?)
>>> incorrectly thinks it isn't, and you propose to add dead code to make
>>> the checker happy?
>>>
>>> I disagree with this approach. Ideally the checker must be improved to
>>
>> Me too. I understand and accept that we sometimes initialize variables
>> to make he compiler happy, but this goes a bit too far. We really should
>> not add dead code - it creates the impression that it can be reached,
>> and would live forever for no good reason.
>>
>>> understand that the code is correct. If that's not possible, we should
>>> be allowed to annotate the code to skip that specific check on these
>>> specific lines, because it has been inspected by a knowledgeable human
>>> and confirmed to be correct.
>>>
>> Agreed.
>>
>>> And if that it still not "possible", then the least intrusive fix would > be to make one of the valid cases the default. But adding new code
>>> which will never be executed, but must still be compiled and stored,
>>> no, thank you. Another code checker could legitimately complain about
>>> that actually.
>>>
>>> IMHO if code checkers return false positives then they are not helping
>>> us and should not be used in the first place.
>>>
>> Checkers are always only providing guidelines and should never be taken
>> at face value.
>>
>> In summary - NACK.
>>
>> Guenter
>>
> 
> What you guys repeatedly called "checker" is GCC 8.
> 
> Intel's 0day bot reported this, and I was also able to reproduce the warnings
> by using the kernel.org toolchain available at:
> https://mirrors.edge.kernel.org/pub/tools/crosstool/files/bin/x86_64/8.1.0/x86_64-gcc-8.1.0-nolibc-sh4-linux.tar.xz
> 
> 
> I also checked "git log --grep=Warray-bounds",
> and I saw people were fixing this kind of warnings.
> And, I am really annoyed by the 0day bot.
> 
> That's why I sent this patch
> despite I have no interest in this driver.
> 
> 
> Having said that, I cannot reproduce these warnings
> by other compilers than sh4-linux-gcc.
> 
> So, probably these warnings are false positive.
> 
> 
> 
> Currently, I have 3 options I can do:
> 
> [1]  I will send an alternative patch to
>       clarify the unreachable path for both compilers and humans
>       without adding dead code.
> 
>    |@@ -351,6 +351,8 @@ static ssize_t fan_div_store(struct device *dev,
>    |                tmp |= data->fan_div[2] << 4;
>    |                smsc47m1_write_value(data, SMSC47M2_REG_FANDIV3, tmp);
>    |                break;
>    |+       default:
>    |+               unreachable();
>    |        }
>    |
>    |        /* Preserve fan min */
> 
> [2] I will send your feed-backs to the maintainer of 0day bot,
>      and persuade him to stop reporting this.
> 
> [3] I will accept that 0day bot will continue sending this report forever.
>      So, I will configure my mailer so that this report
>      will immediately go to the trash box.
> 
> 
> 
> 
> As I already said, I have _zero_ interest in this driver.
> 
> Only the problem I have is, I repeatedly receive annoying reports
> from 0day bot, where my patch is not the root cause of the warnings at all.
> 
I know, happens all the time to me as well. Worse is that in some cases
I don't even know how to fix it (it is not always as easy as it is here).

> Looks like you both believe the current code is OK as-is,
> so I am not sure you are happy with [1].
> 
I am not exactly happy about it, but I would accept it. Just add to the
description that this is to make the compiler happy.

Guenter

> I can try [2] first.
> 
> [3] is the last resort for me.
> 
>