Re: [PATCH] cpuset export symbols gpl

[Robin Holt <holt@sgi.com>]

On Wed, Nov 16, 2005 at 08:16:27AM +0000, Christoph Hellwig wrote:
> On Tue, Nov 15, 2005 at 06:03:36PM -0800, Paul Jackson wrote:
> > Andrew wrote (of exporting cpuset symbols)
> > > We normally would do this when such modules are merged.  Do tell us more..
> > 
> > It was an oversight not to do this when cpusets went in last year,
> > but we didn't notice, as the loadable module we cared about had a
> > hack in place from earlier development that avoided needing this.
> > 
> > In cleaning this up, we realized that the module needed to access
> > task->cpuset->cpus_allowed, and that the correct (and safe) way to
> > do this, via cpuset_cpus_allowed(), was not available to the module.
> > 
> > The other 4 exports I added on general principles, but don't have
> > any pressing need for.  The one I need is cpuset_cpus_allowed().
> > 
> > The loadable module in question we call 'dplace', and is used to
> > provide fancier cpuset-relative task placement by manipulating
> > task->cpus_allowed at exec.
> 
> Again, where is the module.  Please submit the change to export the
> symbols in the same patch series as that module.  And honestly I don't
> think it'll survive review when it's poking that deeply into cpuset
> internals, but we'll see how to do it properly once it's sent here.

I would argue that it does not dig deeply enough.  I think it would be
better to expand dplace to handle early-for activity.  It would be nice
to get a hook in do_fork before the call to copy_process so we can place
the child task struct on the destination instead of the source node.

That said, I could swear that dplace or something that looks a lot like
dplace was already posted on lkml, but I did not find it when I searched.
Maybe my archive is missing some stuff.

Robin