Re: [PATCH 1/3] radix-tree: propagate all tags in idr tree

[Matthew Wilcox]

On Tue, Jun 28, 2022 at 10:53:50AM -0400, Brian Foster wrote:
> On Tue, Jun 28, 2022 at 02:55:11PM +0200, Christian Brauner wrote:
> > On Wed, Jun 15, 2022 at 05:34:02PM +0100, Matthew Wilcox wrote:
> > > On Wed, Jun 15, 2022 at 10:43:33AM -0400, Brian Foster wrote:
> > > > Interesting, thanks. I'll have to dig more into this to grok the current
> > > > state of the radix-tree interface vs. the underlying data structure. If
> > > > I follow correctly, you're saying the radix-tree api is essentially
> > > > already a translation layer to the xarray these days, and we just need
> > > > to move legacy users off the radix-tree api so we can eventually kill it
> > > > off...
> > > 
> > > If only it were that easy ... the XArray has a whole bunch of debugging
> > > asserts to make sure the users are actually using it correctly, and a
> > > lot of radix tree users don't (they're probably not buggy, but they
> > > don't use the XArray's embedded lock).
> > > 
> > > Anyway, here's a first cut at converting the PID allocator from the IDR
> > > to the XArray API.  It boots, but I haven't tried to do anything tricky
> > > with PID namespaces or CRIU.
> > 
> > It'd be great to see that conversion done.
> > Fwiw, there's test cases for e.g. nested pid namespace creation with
> > specifically requested PIDs in
> > 
> 
> Ok, but I'm a little confused. Why open code the xarray usage as opposed
> to work the idr bits closer to being able to use the xarray api (and/or
> work the xarray to better support the idr use case)? I see 150+ callers
> of idr_init(). Is the goal to eventually open code them all? That seems
> a lot of potential api churn for something that is presumably a generic
> interface (and perhaps inconsistent with ida, which looks like it uses
> xarray directly?), but I'm probably missing details.

It's not "open coding".  It's "using the XArray API instead of the
IDR API".  The IDR API is inferior in a number of ways, and yes, I
do want to be rid of it entirely.