[PATCH v4 0/2] workqueue: Make WQ_UNBOUND the default behavior

[PATCH v4 0/2] workqueue: Make WQ_UNBOUND the default behavior

[Marco Crivellari]

Hi,

Currently alloc_workqueue() make sure one among WQ_UNBOUND or WQ_PERCPU is
present. This is enforced warning and setting WQ_PERCPU as default when
alloc_workqueue() without WQ_PERCPU or WQ_UNBOUND.

[ P1 ] change this behavior: if no flags are specified, WQ_UNBOUND is
set by default, changing the default behavior when flags are not present.

WQ_BH is always per-CPU, so WQ_PERCPU is forced in case it is not present.

The WARN_ONCE() when neither flags are present has been removed and it has
been replaced by pr_warn_once(): from now on, every alloc_users()
without WQ_PERCPU will be considered unbound.

Q: I decided to pr_warn_once() just to make sure users are aware of the new
behavior. It makes sense or we can just drop it directly?
I don't think the WARN_ONCE() is appropriate now, for this specific path.

[ P2 ] updates the documentation accordingly.

Thanks!

---
Changes in v4:
- WQ_BH: add documentation about BH workqueue implying WQ_PERCPU

Link to v3: https://lore.kernel.org/all/20260904141421.287647-1-marco.crivellari@suse.com/

Changes in v3:
- WQ_BH require WQ_PERCPU

Link to v2: https://lore.kernel.org/all/20260903165251.481362-1-marco.crivellari@suse.com/

Changes in v2:
- if/else to add/drop WQ_UNBOUND/WQ_PERCPU moved before wq_size assignment

Link to v1: https://lore.kernel.org/all/20260903152204.401638-1-marco.crivellari@suse.com/

Marco Crivellari (2):
  workqueue: Make alloc_workqueue() unbound by default
  doc: Update WQ_UNBOUND and WQ_BH documentation

 Documentation/core-api/workqueue.rst |  9 ++++--
 kernel/workqueue.c                   | 44 ++++++++++++++++++----------
 2 files changed, 35 insertions(+), 18 deletions(-)

-- 
2.55.0

[PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Marco Crivellari]

Currently alloc_workqueue() ensure that one among WQ_PERCPU or WQ_UNBOUND
is present. Now every user has one among these two flags.

In order to achive the future goal of removing WQ_UNBOUND from the code,
start changing the default behavior making alloc_workqueue() act this way:

  - if WQ_PERCPU and WQ_UNBOUND are present: remove WQ_PERCPU (as before)
  - if neither are present: add WQ_UNBOUND

BH workqueues (WQ_BH) are always per-CPU and can't be unbound, so
WQ_PERCPU is now unconditionally forced for them before the above
defaulting logic runs.

The WARN_ONCE() when neither flags are present has been removed and it has
been replaced by pr_warn_once(): from now on, every alloc_users()
without WQ_PERCPU will be considered unbound.

The pr_warn_once() will be removed along with WQ_UNBOUND in the future.

Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 kernel/workqueue.c | 44 ++++++++++++++++++++++++++++----------------
 1 file changed, 28 insertions(+), 16 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index c56c042d1c35..aba94a8e0c38 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -5895,12 +5895,33 @@ static struct workqueue_struct *__alloc_workqueue(const char *fmt,
 	struct workqueue_struct *wq;
 	size_t wq_size;
 	int name_len;
+	bool wq_unbound_by_default = false;
+	bool wq_dropped_percpu = false;
 
 	if (flags & WQ_BH) {
 		if (WARN_ON_ONCE(flags & ~__WQ_BH_ALLOWS))
 			return NULL;
 		if (WARN_ON_ONCE(max_active))
 			return NULL;
+		flags |= WQ_PERCPU;
+	}
+
+	/*
+	 * Every caller that doesn't explicitly specify WQ_PERCPU will be
+	 * unbound (WQ_UNBOUND) by default.
+	 *
+	 * If both flags are present at the same time, WQ_PERCPU will be
+	 * removed.
+	 *
+	 * This must happen before wq_size is computed below, since that
+	 * depends on the finalized WQ_UNBOUND flag.
+	 */
+	if (unlikely(!(flags & (WQ_UNBOUND | WQ_PERCPU)))) {
+		flags |= WQ_UNBOUND;
+		wq_unbound_by_default = true;
+	} else if (unlikely((flags & WQ_PERCPU) && (flags & WQ_UNBOUND))) {
+		flags &= ~WQ_PERCPU;
+		wq_dropped_percpu = true;
 	}
 
 	/* see the comment above the definition of WQ_POWER_EFFICIENT */
@@ -5927,22 +5948,13 @@ static struct workqueue_struct *__alloc_workqueue(const char *fmt,
 		pr_warn_once("workqueue: name exceeds WQ_NAME_LEN. Truncating to: %s\n",
 			     wq->name);
 
-	/*
-	 * One among WQ_PERCPU and WQ_UNBOUND must be set, but not both.
-	 * - If neither is set, default to WQ_PERCPU
-	 * - If both are set, default to WQ_UNBOUND
-	 *
-	 * This code can be removed after workqueue are unbound by default
-	 */
-	if (unlikely(!(flags & (WQ_UNBOUND | WQ_PERCPU)))) {
-		WARN_ONCE(1, "workqueue: %s is using neither WQ_PERCPU or WQ_UNBOUND. "
-			  "Setting WQ_PERCPU.\n", wq->name);
-		flags |= WQ_PERCPU;
-	} else if (unlikely((flags & WQ_PERCPU) && (flags & WQ_UNBOUND))) {
-		WARN_ONCE(1, "workqueue: %s uses both WQ_PERCPU and WQ_UNBOUND. "
-			  "Dropped WQ_PERCPU, keeping WQ_UNBOUND.\n", wq->name);
-		flags &= ~WQ_PERCPU;
-	}
+	if (unlikely(wq_unbound_by_default))
+		pr_warn_once("workqueue: %s is using neither WQ_PERCPU or WQ_UNBOUND. "
+			     "Setting WQ_UNBOUND.\n", wq->name);
+
+	WARN_ONCE(wq_dropped_percpu,
+		  "workqueue: %s uses both WQ_PERCPU and WQ_UNBOUND. "
+		  "Dropped WQ_PERCPU, keeping WQ_UNBOUND.\n", wq->name);
 
 	if (flags & WQ_BH) {
 		/*
-- 
2.55.0

[PATCH v4 2/2] doc: Update WQ_UNBOUND and WQ_BH documentation

[Marco Crivellari]

When WQ_PERCPU is not present the default behavior is now WQ_UNBOUND,
so update the documentation, reflecting this change.

WQ_BH workqueues are the exception to this rule: they are always
per-CPU and can't be unbound, so WQ_PERCPU is implicitly set for them
regardless of what the caller passes. Update the WQ_BH section
accordingly, as WQ_HIGHPRI is no longer the only additional flag
involved.

Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 Documentation/core-api/workqueue.rst | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/Documentation/core-api/workqueue.rst b/Documentation/core-api/workqueue.rst
index bb770f556568..76b70e7f4e3f 100644
--- a/Documentation/core-api/workqueue.rst
+++ b/Documentation/core-api/workqueue.rst
@@ -177,8 +177,9 @@ resources, scheduled and executed.
   workqueues are always per-CPU and all BH work items are executed in the
   queueing CPU's softirq context in the queueing order.
 
-  All BH workqueues must have 0 ``max_active`` and ``WQ_HIGHPRI`` is the
-  only allowed additional flag.
+  All BH workqueues must have 0 ``max_active``. ``WQ_HIGHPRI`` and
+  ``WQ_PERCPU`` are the only permitted flags. ``WQ_PERCPU`` is enforced by
+  default as BH workqueueus cannot be unbound.
 
   BH work items cannot sleep. All other features such as delayed queueing,
   flushing and canceling are supported.
@@ -206,6 +207,10 @@ resources, scheduled and executed.
   * Long running CPU intensive workloads which can be better
     managed by the system scheduler.
 
+  Workqueue without ``WQ_PERCPU`` are unbound by default, with the
+  exception of ``WQ_BH`` workqueues, which always have ``WQ_PERCPU``
+  set since they can't be unbound.
+
 ``WQ_FREEZABLE``
   A freezable wq participates in the freeze phase of the system
   suspend operations.  Work items on the wq are drained and no
-- 
2.55.0

Re: [PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Breno Leitao]

On Thu, Oct 15, 2026 at 04:01:47PM +0200, Marco Crivellari wrote:
> +	if (unlikely(wq_unbound_by_default))
> +		pr_warn_once("workqueue: %s is using neither WQ_PERCPU or WQ_UNBOUND. "
> +			     "Setting WQ_UNBOUND.\n", wq->name);

Why do we need to warn here? With this new approach, shouldn't users be
able to rely on WQ_UNBOUND being the default and leave it out?

Thanks,
--breno

Re: [PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Marco Crivellari]

On Fri, Oct 16, 2026 at 1:45 PM Breno Leitao <leitao@debian.org> wrote:
>
> On Thu, Oct 15, 2026 at 04:01:47PM +0200, Marco Crivellari wrote:
> > +     if (unlikely(wq_unbound_by_default))
> > +             pr_warn_once("workqueue: %s is using neither WQ_PERCPU or WQ_UNBOUND. "
> > +                          "Setting WQ_UNBOUND.\n", wq->name);
>
> Why do we need to warn here? With this new approach, shouldn't users be
> able to rely on WQ_UNBOUND being the default and leave it out?

Yes, it was mostly to ensure users notice the change in behavior when
flags are not specified.

But I can remove it for the next version.

Thanks!

-- 

Marco Crivellari

SUSE Labs

Re: [PATCH v4 2/2] doc: Update WQ_UNBOUND and WQ_BH documentation

[Breno Leitao]

On Thu, Oct 15, 2026 at 04:01:48PM +0200, Marco Crivellari wrote:
> When WQ_PERCPU is not present the default behavior is now WQ_UNBOUND,
> so update the documentation, reflecting this change.
> 
> WQ_BH workqueues are the exception to this rule: they are always
> per-CPU and can't be unbound, so WQ_PERCPU is implicitly set for them
> regardless of what the caller passes. Update the WQ_BH section
> accordingly, as WQ_HIGHPRI is no longer the only additional flag
> involved.
> 
> Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
> Suggested-by: Tejun Heo <tj@kernel.org>
> Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>

Reviewed-by: Breno Leitao <leitao@debian.org>

Re: [PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Breno Leitao]

On Thu, Oct 15, 2026 at 04:01:47PM +0200, Marco Crivellari wrote:
> Currently alloc_workqueue() ensure that one among WQ_PERCPU or WQ_UNBOUND
> is present. Now every user has one among these two flags.
> 
> In order to achive the future goal of removing WQ_UNBOUND from the code,
> start changing the default behavior making alloc_workqueue() act this way:
> 
>   - if WQ_PERCPU and WQ_UNBOUND are present: remove WQ_PERCPU (as before)
>   - if neither are present: add WQ_UNBOUND
> 
> BH workqueues (WQ_BH) are always per-CPU and can't be unbound, so
> WQ_PERCPU is now unconditionally forced for them before the above
> defaulting logic runs.
> 
> The WARN_ONCE() when neither flags are present has been removed and it has
> been replaced by pr_warn_once(): from now on, every alloc_users()
> without WQ_PERCPU will be considered unbound.
> 
> The pr_warn_once() will be removed along with WQ_UNBOUND in the future.
> 
> Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
> Suggested-by: Tejun Heo <tj@kernel.org>
> Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>

Reviewed-by: Breno Leitao <leitao@debian.org>

> @@ -5895,12 +5895,33 @@ static struct workqueue_struct *__alloc_workqueue(const char *fmt,
>  	struct workqueue_struct *wq;
>  	size_t wq_size;
>  	int name_len;
> +	bool wq_unbound_by_default = false;
> +	bool wq_dropped_percpu = false;

Nit: I think these declarations would look better in reverse xmas tree
order. I know Tejun doesn't feel strongly about it, so it's just a nit.

--breno

Re: [PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Marco Crivellari]

On Fri, Oct 16, 2026 at 5:54 PM Breno Leitao <leitao@debian.org> wrote:
> [...]
> > @@ -5895,12 +5895,33 @@ static struct workqueue_struct *__alloc_workqueue(const char *fmt,
> >       struct workqueue_struct *wq;
> >       size_t wq_size;
> >       int name_len;
> > +     bool wq_unbound_by_default = false;
> > +     bool wq_dropped_percpu = false;
>
> Nit: I think these declarations would look better in reverse xmas tree
> order. I know Tejun doesn't feel strongly about it, so it's just a nit.
>
> --breno

Cool, thanks for the hint, I will change this along with the removal
of pr_warn_once().

-- 

Marco Crivellari

SUSE Labs

Re: [PATCH v4 1/2] workqueue: Make alloc_workqueue() unbound by default

[Marco Crivellari]

On Fri, Oct 16, 2026 at 6:18 PM Marco Crivellari
<marco.crivellari@suse.com> wrote:
>
> On Fri, Oct 16, 2026 at 5:54 PM Breno Leitao <leitao@debian.org> wrote:
> > [...]
> > > @@ -5895,12 +5895,33 @@ static struct workqueue_struct *__alloc_workqueue(const char *fmt,
> > >       struct workqueue_struct *wq;
> > >       size_t wq_size;
> > >       int name_len;
> > > +     bool wq_unbound_by_default = false;
> > > +     bool wq_dropped_percpu = false;
> >
> > Nit: I think these declarations would look better in reverse xmas tree
> > order. I know Tejun doesn't feel strongly about it, so it's just a nit.
> >
> > --breno
>
> Cool, thanks for the hint, I will change this along with the removal
> of pr_warn_once().
>

Oh wait: only one of the bool, the other is going away with the
removal of the pr_warn_once().

-- 

Marco Crivellari

SUSE Labs

Re: [PATCH v4 0/2] workqueue: Make WQ_UNBOUND the default behavior

[Tejun Heo]

Hello, Marco.

The following is a Claude-generated review.

On Thu, Oct 15, 2026 at 04:01:46PM +0200, Marco Crivellari wrote:
> Currently alloc_workqueue() make sure one among WQ_UNBOUND or WQ_PERCPU is
> present. This is enforced warning and setting WQ_PERCPU as default when
> alloc_workqueue() without WQ_PERCPU or WQ_UNBOUND.

- 1/2: The description gives the future removal of WQ_UNBOUND as the only
  reason. Can it lead with why unbound should be the default, as
  dadb3ebcf395 ("workqueue: WQ_PERCPU added to alloc_workqueue users")
  does?

  Moving the defaulting above the wq_size computation is the one subtle
  point. Maybe add a sentence on it to the description?

  The new comment says WQ_UNBOUND is final at that point, but the
  WQ_POWER_EFFICIENT block right below can still set it. The first two
  paragraphs restate the if/else. Maybe trim it to the ordering
  constraint against wq_size?

  unlikely() on the neither-flag branch now marks the path that's meant
  to become the default. Maybe drop it?

  The WQ_POWER_EFFICIENT comment in include/linux/workqueue.h says such
  workqueues are per-cpu by default. With this patch, one without
  WQ_PERCPU is unbound regardless of workqueue.power_efficient. All
  in-tree users pass WQ_PERCPU, so only the text needs updating.

  Typos: "achive", "alloc_users()" (alloc_workqueue() users), "ensure" ->
  "ensures", "neither flags are" -> "neither flag is".

- 2/2: The description says WQ_PERCPU is set for WQ_BH "regardless of
  what the caller passes", but WQ_BH | WQ_UNBOUND is rejected by the
  __WQ_BH_ALLOWS check.

  "the only permitted flags" dropped "additional", but WQ_BH itself is
  in __WQ_BH_ALLOWS too. Also, "workqueueus" typo.

  The new WQ_UNBOUND sentence sits after the "useful for the following
  cases" bullets and reads as one more case. Maybe move it to the top of
  the entry? Also, "Workqueue without" -> "Workqueues without".

  The Guidelines section still says "using a bound wq is usually
  beneficial", which now contradicts the default. Can that be updated
  too?

  Nit: doc patches in this tree usually use the "workqueue: doc:" prefix.

Thanks.

-- 
tejun