[Devel] Re: [-mm] CPU controller statistics (v5)

[Paul Menage]

On Thu, Jun 5, 2008 at 2:37 AM, Balbir Singh <balbir at linux.vnet.ibm.com> wrote:
>>
>> Can we avoid these tests?  By requiring that cgroup_subsys.initialize()
>> always be non-zero?  It might make sense, and it might not...
>>
>
> They are really hard to avoid, otherwise we might be taking away the flexibility
> we have.

And this is something that only gets called at startup.

>
>> Also, if this code is likely to be executed with any frequency then the
>> test of `early' could be inlined:
>>
>> static inline void cpu_cgroup_initialize(int early)
>> {
>>       if (unlikely(!early))
>>               __cpu_cgroup_initialize();
>> }
>>
>> yes?
>>
>
> Definitely

Er, no. It gets called twice at system boot (once with early=true and
once with early=false) via a vtable. So there's nothing to optimize
for, and making the function inline won't help since the compiler
needs to take its address.

Paul
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers