Re: [PATCH 2/2] workqueue: implement lockup detector

[Don Zickus <dzickus@redhat.com>]

On Fri, Dec 04, 2015 at 09:02:26AM +0100, Ingo Molnar wrote:
> 
> * Tejun Heo <tj@kernel.org> wrote:
> 
> > Hello, Ulrich.
> > 
> > On Thu, Dec 03, 2015 at 03:12:20PM -0500, Ulrich Obergfell wrote:
> > > I share Don's concern about connecting the soft lockup detector and the
> > > workqueue watchdog to the same kernel parameter in /proc. I would feel
> > > more comfortable if the workqueue watchdog had its dedicated parameter.
> > 
> > Sure, separating the knobs out isn't difficult.  I still don't like
> > the idea of having multiple set of similar knobs controlling about the
> > same thing tho.
> > 
> > For example, let's say there's a user who boots with "nosoftlockup"
> > explicitly.  I'm pretty sure the user wouldn't be intending to keep
> > workqueue watchdog running.  The same goes for threshold adjustments,
> > so here's my question.  What are the reasons for the concern?  What
> > are we worrying about?
> 
> As Don mentioned it already, we went through similar arguments (and pain) with the 
> hard/soft lockup detectors and its various control knobs, it would be better to 
> have new control knobs separated.
> 
> As for the ease of use argument, we can add a new, obviously named control knob 
> that controls _all_ lockup detectors:
> 
>   boot param: nolockupdetectors
>   matching Kconfig knob: CONFIG_BOOTPARAM_NO_LOCKUP_DETECTORS=0
> 
> but please don't artificially couple the control knobs of these various lockup 
> detectors, as these internal assumptions are less than obvious to users. With 
> (effectively) 4 lockup detectors such coupling of interfaces is even more 
> confusing and damaging.

It might be worth tying them together with a generic knob and expanding the
bit mask for the 'watchdog' variable.  I can't figure out an easy way to do
that right now.

I don't think we want to go down the route of 'registering' detectors yet.
:-)

Cheers,
Don