On Mon, Sep 20, 2010 at 11:26:47PM +0200, Tejun Heo wrote:

Hello,

On 09/20/2010 10:35 PM, Roland Dreier wrote:

quoted

Looks fine to me as an improvement over the status quo, but I wonder how
many of these places could use the radix_tree stuff instead?  If you're
not using the ability of the idr code to assign an id for you, then it
seems the radix_tree API is a better fit.

I agree.  Wouldn't those users better off simply using radix tree?

It could go either way. I was about to write the same function when
playing with it for IRQ mapping, the idea being to propagate the initial
tree with sparse static vectors and then switch over to dynamic IDs for
virtual IRQ creation. I ended up going with a radix tree for other
reasons, though.