Re: [PATCH -v3] KVM: x86: lapic: Clean up find_highest_vector() and count_vectors()

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Sep 05, 2012 at 06:40:26PM +0900, Takuya Yoshikawa wrote:
> On Wed, 5 Sep 2012 12:26:49 +0300
> "Michael S. Tsirkin" <mst@redhat.com> wrote:
> 
> > It's not guaranteed if another thread can modify the bitmap.
> > Is this the case here? If yes we need at least ACCESS_ONCE.
> 
> In this patch, using the wrapper function to read out a register
> value forces compilers not to do bad things.

It's not robust. Compiler is free to optimize it out, and it
frequently does that.

> But I agree that it is not a good API.
> 
> I would like to use fls() rather than ACCESS_ONCE if it's
> really needed.

Sure, makes sense.

> > > Anyway, I'm now thinking that we do not care about such
> > > things here, and can just follow your advice, yes?
> > 
> > Unless you see an issue with it ...
> 
> Although I read the code, I'm not sure.
> 
> But this code is apparently not so critical for performance
> that we can simply use fls().

Yes. I guess if it becomes critical we'll need to add a cache anyway.

> If I can remove likely() and use proper macros, that's enough
> for me.

Me too.

> Thanks,
> 	Takuya