Re: [PATCH v6 0/2] workqueue: Make WQ_UNBOUND the default behavior

From: Frederic Weisbecker

Date: Wed Nov 18 2026 - 09:54:14 EST


Le Mon, Nov 16, 2026 at 04:50:49PM +0100, Marco Crivellari a écrit :
> Hi,
>
> Currently alloc_workqueue() make sure one among WQ_UNBOUND or WQ_PERCPU is
> present with the current default being WQ_PERCPU.
>
> Per-CPU workqueues provide strict CPU affinity and concurrency
> management. Some users rely on these properties, but many don't need
> them, and for those users per-CPU workqueues significantly restrict
> what the scheduler can do with their work items.
>
> Unbound workqueues now use non-strict (soft) affinity scopes, so they
> keep part of the locality benefits while giving the scheduler more
> flexibility. This helps system-level concerns such as power management
> and CPU isolation.
>
> This series therefore transitions alloc_workqueue() to unbound as the
> default. Users that need per-CPU behavior can still request it
> explicitly with WQ_PERCPU.
>
> The behavior changes as follows:
>
> [ P1 ]
>
> - keep WARN_ON() if WQ_PERCPU and WQ_UNBOUND are present and drop
> WQ_PERCPU.
>
> - defaults to WQ_UNBOUND except for WQ_BH, that it is always per-CPU.
>
> - WQ_POWER_EFFICIENT description, kernel parameters and konfig
> description changed to reflect what this patch introduces.
>
> [ P2 ]
>
> Documentation updated accordingly.
>
> Thanks!

That's a big step forward.

Thanks a lot!

--
Frederic Weisbecker
SUSE Labs