Re: [PATCH] gpio: aspeed: fix compile testing warning

[Arnd Bergmann]

On Mon, Jul 9, 2018 at 5:31 PM, Alexander Stein
<alexander.stein@xxxxxxxxxxxxxxxxxxxxx> wrote:
> On Monday, July 9, 2018, 4:56:03 PM CEST Arnd Bergmann wrote:
>> Gcc cannot always see that BUG_ON(1) is guaranteed to not
>> return, so we get a warning message in some configurations:
>>
>> drivers/gpio/gpio-aspeed.c: In function 'bank_reg':
>> drivers/gpio/gpio-aspeed.c:244:1: error: control reaches end of non-void
>> function [-Werror=return-type]
>>
>> Using a plain BUG() is easier here and avoids the problem.
>>
>> Fixes: 44ddf559d579 ("gpio: aspeed: Rework register type accessors")
>> Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
>> ---
>>  drivers/gpio/gpio-aspeed.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpio/gpio-aspeed.c b/drivers/gpio/gpio-aspeed.c
>> index 1e00f4045f9d..2342e154029b 100644
>> --- a/drivers/gpio/gpio-aspeed.c
>> +++ b/drivers/gpio/gpio-aspeed.c
>> @@ -240,7 +240,7 @@ static inline void __iomem *bank_reg(struct aspeed_gpio
>> *gpio, case reg_cmdsrc1:
>>               return gpio->base + bank->cmdsrc_regs + GPIO_CMDSRC_1;
>>       }
>> -     BUG_ON(1);
>> +     BUG();
>>  }
>>
>>  #define GPIO_BANK(x) ((x) >> 5)
>
> Is the semantic of BUG() (and BUG_ON as well) to never return?

On most architectures and configurations yes, but not on some of
the minor architectures if CONFIG_BUG is disabled.

> If so, then
> just an idea: Is it possible to add some macro magic in BUG_ON(x) to fail
> compiling if x is compile-constant? Giving a hint the passed condition always
> fails, which indicates a problem, at least to me.

Not sure, that might not work well in cases where it's a compile-time
constant in some configurations but variable in others.

> From a short search I found this in drivers/gpu/vga/vgaarb.c L630-633:
>>       if (vgadev_find(pdev) != NULL) {
>>               BUG_ON(1);
>>               goto fail;
>>       }
> You can't fail with a BUG_ON(1) and try to do some error handling after that.

Right.

Traditionally when CONFIG_BUG was disabled, we would have continued
here, so that could have been intentional, but in any case a WARN_ON()
would have been more appropriate here.

      Arnd