On 11/12/2012 10:23 AM, Russell King - ARM Linux wrote:
> On Mon, Nov 12, 2012 at 04:12:29PM +0100, Maarten Lankhorst wrote:
>> Op 08-11-12 21:23, Sasha Levin schreef:
>>> @@ -465,10 +465,8 @@ static void __init combiner_cascade_irq(unsigned int combiner_nr, unsigned int i
>>> else
>>> max_nr = EXYNOS4_MAX_COMBINER_NR;
>>>
>>> - if (combiner_nr >= max_nr)
>>> - BUG();
>>> - if (irq_set_handler_data(irq, &combiner_data[combiner_nr]) != 0)
>>> - BUG();
>>> + BUG_ON(combiner_nr >= max_nr);
>>> + BUG_ON(irq_set_handler_data(irq, &combiner_data[combiner_nr]) != 0);
>>
>> Is it really a good idea to put functions that perform work in a BUG_ON?
>> I don't know, but for some reason it just feels wrong. I'd expect code to
>> compile fine if BUG_ON was a noop, so doing verification calls only, not
>> actual work..
>
> Well, it is currently defined as:
>
> include/asm-generic/bug.h:#define BUG_ON(condition) do { if (unlikely(condition)) BUG(); } while(0)
> include/asm-generic/bug.h:#define BUG_ON(condition) do { if (condition) ; } while(0)
>
> but as these can be overridden, I don't think relying on those
> implementations is a good idea; to do so would be fragile. Eg, what if
> the BUG_ON() implementation becomes just:
>
> #define BUG_ON(x)
>
> then the function call itself vanishes. So, only put the actual bug test
> inside a BUG_ON(), not the functional part which must always be executed.
Even if we ignore that modifying the side-effects is wrong, there's already
more than enough code in the kernel (both in kernel/ / mm/, and in arch/) to
cause breakage if for some reason the expression is not evaluated.
If some arch decides to not evaluate the expression there it's going to be
inherently broken.
Thanks,
Sasha
--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
