system_nrt_wq, system suspend, and the freezer
jlayton at samba.org
Thu Feb 16 10:59:45 PST 2012
On Thu, 16 Feb 2012 08:29:51 -0800
Tejun Heo <tj at kernel.org> wrote:
> On Thu, Feb 16, 2012 at 10:27:28AM -0500, Jeff Layton wrote:
> > These should all be freezable and we might even be able to get away
> > with WQ_UNBOUND for some of these.
> In general, I would recommend specifying as few special attribute as
> possible. If WQ_UNBOUND is necessary (large amount of CPU cycles
> consumed, extremely high concurrency), sure, but I think we're
> generally better off using as default attributes as possible. It just
> makes things much easier later when we need to implement new features
> or update the implementation.
Ok, fair enough. Probably no need to make it unbound...
> > I think we put most of these in system_nrt_wq because Tejun put an
> > earlier job into that queue when he converted it from slow_work and we
> > just cargo-cult copied that...
> > I'll spend some time looking at this in the next day or two, but I
> > suspect that the right answer is to just move these off of the "public"
> > workqueues altogether.
> If freezing & nrt is everything necessary, just create
> system_nrt_freezable_wq and use that.
The other problem here is that we really ought to be submitting the
write completion handler to a workqueue that has WQ_MEM_RECLAIM set.
Since none of the public wq's have that then I guess we'll have to make
Jeff Layton <jlayton at samba.org>
More information about the dri-devel