Re: [PATCH] x86: Replace uses of current_cpu_data with this_cpu ops

[Tejun Heo <tj@kernel.org>]

Hello,

On 12/08/2010 06:17 PM, Eric Dumazet wrote:
> Le mercredi 08 décembre 2010 à 09:33 -0600, Christoph Lameter a écrit :
>> On Wed, 8 Dec 2010, Eric Dumazet wrote:
>>
>>>> In the long run, it might be a good idea to remove cpu_data() macro
>>>> too and use per_cpu macro directly.
>>>>
>>>
>>> or introduce this_cpu_has() to remove the adress computation
>>>
>>> -	if (cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_ARAT)) {
>>>
>>> +	if (this_cpu_has(X86_FEATURE_ARAT)) {
>>>
>>
>> The fundamental problem is that bitops require memory addresses which does
>> not work with per cpu ops.
>>
> 
> Sure, set() or clear() needs the address, but the read doesnt ?
> 
> This can be implemented as
> 
>   (__this_cpu_read(cpu_info.word[X]) & MASK)
> 
> Anyway, even if mapped to cpu_has(__this_cpu_ptr(&cpu_info), XXX), 
> it would be cleaner to use this_cpu_has() helper.

Agreed.  I'd also like to see cpu_data() gone too.  It isn't used too
many times and mixed with direct percpu ops only makes things more
confusing without much benefit.  Does somebody care to send patches
for these cleanups?

Thanks.

-- 
tejun