On Thu, 2026-01-08 at 14:45 +0100, Marco Crivellari wrote:
> On Thu, Jan 8, 2026 at 2:29 PM Rodrigo Vivi <[email protected]>
> wrote:
> > I'm sorry for the delay here, but could you please refactor this
> > commit
> > message?
> > 
> > The first part of this commit message is the true justification for
> > your
> > original work, not for this patch here.
> 
> Hi Rodrigo,
> 
> Sure, I already have a new version of it:
> 
> ---
> 
> This patch continues the effort to refactor workqueue APIs, which has
> begun
> with the changes introducing new workqueues and a new alloc_workqueue
> flag:
> 
>    commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and
> system_dfl_wq")
>    commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
> 
> The point of the refactoring is to eventually alter the default
> behavior of
> workqueues to become unbound by default so that their workload
> placement is
> optimized by the scheduler.
> 
> Before that to happen, workqueue users must be converted to the
> better named
> new workqueues with no intended behaviour changes:
> 
>    system_wq -> system_percpu_wq
>    system_unbound_wq -> system_dfl_wq
> 
> This way the old obsolete workqueues (system_wq, system_unbound_wq)
> can be
> removed in the future.
> 
> 
> > Except for your last phrase, which indicates, some wish of removing
> > the unbound_wq, it doesn't state clear on why we should change the
> > unbound per the dfl (default?).
> > 
> > Perhaps the authors of these cases below wanted to be unbound,
> > but choosing the default will make us to be tied to whatever
> > default might become in the future.
> > 
> > Right now both unbound and dfl are identical. In the future
> > you are planning to remove the unbound, but what about the dfl?
> > Any plans or possible changes? If no change is planned to dfl,
> > why create default and simply not stay with the unbound one
> > that is much more clear on its intention?
> 
> Yes, "dfl" is "default".
> 
> With the new version above I think it is clear, but in short: the
> rename
> is needed as a first refactoring step; the plan is to have an unbound
> workqueue
> as default.
> 
> system_unbound_wq will be removed and changed with system_dfl_wq.
> system_dfl_wq is the new unbound workqueue, it works exactly the same
> as
> system_unbound_wq.
> 
> system_wq will be replaced with system_percpu_wq; same thing as
> above,
> no behavioral changes, system_percpu_wq is the new per-cpu workqueue.
> 
> This is to make clear that if a per-cpu workqueue is not needed, the
> correct
> choice is system_dfl_wq.
> 
> Also, during this conversion phase, I can tell many subsystems asked
> to
> change system_wq (the name of the "old" per-cpu workqueue) with
> system_dfl_wq, because they just wanted to use a system workqueue and
> thought system_wq was already the unbound one.
> 
> All of this work has been discussed here:
> 
> https://lore.kernel.org/all/[email protected]/
> 
> I will add this Link to the above commit log. Let me know if it
> sounds good.

This is perfect, thank you!

> 
> Thanks!

Reply via email to