[Devel] [PATCH 1/6] slab: cleanup kmem_cache_create_memcg()

[Pekka Enberg]

On 12/19/2013 11:26 AM, Vasily Averin wrote:
> On 12/19/2013 12:39 PM, Vladimir Davydov wrote:
>> On 12/19/2013 12:17 PM, Vasily Averin wrote:
>>> On 12/18/2013 05:16 PM, Vladimir Davydov wrote:
>>>> --- a/mm/slab_common.c
>>>> +++ b/mm/slab_common.c
>>>> @@ -176,8 +176,9 @@ kmem_cache_create_memcg(struct mem_cgroup *memcg, const char *name, size_t size,
>>>>   	get_online_cpus();
>>>>   	mutex_lock(&slab_mutex);
>>>>   
>>>> -	if (!kmem_cache_sanity_check(memcg, name, size) == 0)
>>>> -		goto out_locked;
>>>> +	err = kmem_cache_sanity_check(memcg, name, size);
>>>> +	if (err)
>>>> +		goto out_unlock;
>>>>   
>>>>   	/*
>>>>   	 * Some allocators will constraint the set of valid flags to a subset
>>> Theoretically in future kmem_cache_sanity_check() can return positive value.
>>> Probably it's better to check (err < 0) in caller ?
>> Hmm, why? What information could positive retval carry here? We have
>> plenty of places throughout the code where we check for (err), not
>> (err<0), simply because it looks clearer, e.g. look at
>> __kmem_cache_create() calls. If it returns a positive value one day, we
>> will have to parse every place where it's called. Anyway, if someone
>> wants to change a function behavior, he must check every place where
>> this function is called and fix them accordingly.
> I believe expected semantic of function -- return negative in case of error.
> So correct error cheek should be (err < 0).
> (err) check is semantically incorrect, and it can lead to troubles in future.

I don't know what semantics you are referring to but a typical 
convention in mm/*.c is to return zero on success and negative on error 
but never positive numbers.

Looking at mm/slab_common.c, "if (err)" is the established convention so 
using "if (err < 0)" just because is pointless here.

                             Pekka