Re: [PATCH 0/5] kmod: Cleanups, simplifications, and make isolation friendly v3

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Jul 27, 2015 at 02:48:40PM -0400, Tejun Heo wrote:
> On Mon, Jul 27, 2015 at 06:27:15PM +0200, Frederic Weisbecker wrote:
> > Hence those two debatable changes:
> > 
> > _ We would like to use generic workqueues. System unbound workqueues are
> >   a very good candidate but they are not wide affine, only node affine.
> >   Now probably a node is enough to perform many parallel kmod jobs.
> 
> If being node-affine is an issue, kmod can easily create a workqueue
> w/o NUMA affinity using apply_workqueue_attrs() with no_numa set to
> %true.

Right, but we would like to get rid of khelper which already plays a
similar role.

> 
> > _ We would like to remove the wait_for_helper kernel thread (UMH_WAIT_PROC
> >   handler) to use the workqueue. It means that if the workqueue blocks,
> >   and no other worker can take pending kmod request, we can be screwed.
> >   Now if we have 512 threads, this should be enough.
> 
> The maximum number of worker can also be raised on the workqueue.
> That said, I don't think we want to.
> 
> IMHO, system_wq should be fine and if it isn't turning off numa
> affinity or raising max worker limit later is pretty trivial.

That's what I think too. How many workers system_unbound_wq can handle? If kmod
raises very high numbers of threads in parallel like > 500, I think that would be
a problem on its own anyway.

Thanks.