Re: [RFC PATCH 0/2] RT scheduling policies for workqueues

[Tejun Heo <tj@kernel.org>]

Hello,

On Mon, Mar 28, 2022 at 12:09:27PM +0200, Marc Kleine-Budde wrote:
> > Having a kthread per "low-latency" tty instance is something I would
> > prefer. The kwork corner is an anonymous worker instance and probably
> > does more harm than good. Especially if it is a knob for everyone which
> > is used for the wrong reasons and manages to be harmful in the end.
> > With a special kthread for a particular tty, the thread can be assigned
> > with the desired priority within the system and ttyS1 can be
> > distinguished from ttyS0 (and so on). This turned out to be useful in a
> > few setups over the years.
> 
> +1
> 
> The networking subsystem has gone the same/similar way with NAPI. NAPI
> handling can be switched from the softirq to kernel thread on a per
> interface basis.

I wonder whether it'd be useful to provide a set of wrappers which can make
switching between workqueue and kworker easy. Semantics-wise, they're
already mostly aligned and it shouldn't be too difficult to e.g. make an
unbounded workqueue be backed by a dedicated kthread_worker instead of
shared pool depending on a flag, or even allow switching dynamically.

Thanks.

-- 
tejun