Re: [PATCH] idr: fix backtrack logic in idr_remove_all

[Imre Deak <imre.deak@nokia.com>]

On Tue, May 18, 2010 at 05:23:20PM +0200, ext Tejun Heo wrote:
> On 05/18/2010 01:18 PM, Imre Deak wrote:
> >> Shouldn't this be id ^ bt_mask?  The above only detects 1 -> 0
> >> transitions not the other way around.
> > 
> > It works according to the following with n=1:
> > 
> > id            id+2            fls((id+2) & ~id)
> > 0             2               2
> > 2             4               3
> > 4             6               2
> > 6             8               4
> > 8             10              2
> > 10            12              3
> > 12            14              2
> > 
> > I think this should work.
> 
> Ah, I thought you were doing fls(id & ~(id + 2)) and thus looking at 1
> -> 0 transitions.  It's the other way and you're looking for the
> highest 0 -> 1 transition which should be the same to the highest bit
> changing if you aren't overflowing.

Yes, both ways you get the same result and in case of overflow neither
will work.

> The patch looks good then.  I still think ^ test would be clearer tho.
> Hmmm...

Well xor results in one instruction less on machines without a nand
instruction, so I'll change that.

> Can you please add little comment there stating that you're
> looking for the highest bit flipping?

Yes, I'll re-test and follow up with an updated patch.

Thanks for the review,
Imre

> 
> Reviewed-by: Tejun Heo <tj@kernel.org>
> 
> Thanks.
>