Re: [PATCH v2 3/4] kthread: NUMA aware kthread_create_on_cpu()

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 10 Dec 2010 08:02:41 +0100 Eric Dumazet <eric.dumazet@gmail.com> wrote:

> Le jeudi 09 d__cembre 2010 __ 22:32 -0800, Andrew Morton a __crit :
> 
> > but but but.  The name "kthread_create_on_cpu" sucks.  It's plain wrong.
> 
> Okay you are right Andrew ;) I dont have better idea for the moment.

Dunno.  kthread_create_with_memory_on_node() :)

How's about kthread_create_for_node()?  That's sufficiently vague to
not mislead readers into thinking that it schedules the thread on that
CPU and leaves room in the namespace for a real kthread_create_on_cpu()
(which we could well end up creating).

kthread_create_node_mem()?

> Note that all callers I converted really create one kthread per cpu, not
> per node. They didnt care of node affinity, only me :)
> 
> kthread_create_on_node() seems misleading to me (some cpus run on
> memoryless nodes)

True, but what we're doing here is specifying on which node the
kthread's memory resources should reside - we need to do that even for
CPUs which live on memoryless nodes.