Re: [thiscpuops upgrade 08/10] percpu: generic this_cpu_cmpxchg() and this_cpu_cmpxchg_double support

[Eric Dumazet <eric.dumazet@gmail.com>]

Le vendredi 26 novembre 2010 à 18:07 +0100, Tejun Heo a écrit :
> On 11/26/2010 06:01 PM, Eric Dumazet wrote:
> >> What I'm afraid of is generic code switching to use it in very hot
> >> paths when a lot of archs can't actually do it leading to performance
> >> regressions.  Is this something which is available on most archs?
> >>
> > 
> > I would say cmpxchg() has the same problem, some arches dont have a
> > native instruction, yet we use it in some paths.
> 
> Yeah, well, there's difference between some not having it and only
> x86_64 having it.  cmpxchg was already rather well received when it
> was added even though it wasn't available on all archs.  I'm not
> against it but think we should use some caution here and think about
> the impact of unoptimized cases which can be pretty common (preemption
> toggling isn't too expensive but irq toggling can be quite).
> 


x86_32 has it too

x86_64 uses cmpxchg16b, while x86_32 uses cmpxhg8b


And I guess it will be used in contexts we had to disable irqs, so it
will be faster on x86, and same cost on other arches.