Currently if a user enqueues a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.

alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.

This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.

This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:

commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.

With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.

Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.

Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
index 6559d72d5d42..9083367ae2e4 100644
--- a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
+++ b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
@@ -255,14 +255,16 @@ static int mdp_probe(struct platform_device *pdev)
 		goto err_free_mutex;
 	}
 
-	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME, WQ_FREEZABLE, 0);
+	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME,
+				      WQ_FREEZABLE | WQ_PERCPU, 0);
 	if (!mdp->job_wq) {
 		dev_err(dev, "Unable to create job workqueue\n");
 		ret = -ENOMEM;
 		goto err_deinit_comp;
 	}
 
-	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock", WQ_FREEZABLE,
+	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock",
+					WQ_FREEZABLE | WQ_PERCPU,
 					0);
 	if (!mdp->clock_wq) {
 		dev_err(dev, "Unable to create clock workqueue\n");
-- 
2.51.1

Il 07/11/25 15:13, Marco Crivellari ha scritto:
> Currently if a user enqueues a work item using schedule_delayed_work() the
> used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
> WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
> schedule_work() that is using system_wq and queue_work(), that makes use
> again of WORK_CPU_UNBOUND.
> This lack of consistency cannot be addressed without refactoring the API.
> 
> alloc_workqueue() treats all queues as per-CPU by default, while unbound
> workqueues must opt-in via WQ_UNBOUND.
> 
> This default is suboptimal: most workloads benefit from unbound queues,
> allowing the scheduler to place worker threads where they’re needed and
> reducing noise when CPUs are isolated.
> 
> This continues the effort to refactor workqueue APIs, which began with
> the introduction of new workqueues and a new alloc_workqueue flag in:
> 
> commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
> commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
> 
> This change adds a new WQ_PERCPU flag to explicitly request
> alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
> 
> With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
> any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
> must now use WQ_PERCPU.
> 
> Once migration is complete, WQ_UNBOUND can be removed and unbound will
> become the implicit default.
> 
> Suggested-by: Tejun Heo <tj@kernel.org>
> Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>

Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>

> ---
>   drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> index 6559d72d5d42..9083367ae2e4 100644
> --- a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> +++ b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> @@ -255,14 +255,16 @@ static int mdp_probe(struct platform_device *pdev)
>   		goto err_free_mutex;
>   	}
>   
> -	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME, WQ_FREEZABLE, 0);
> +	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME,
> +				      WQ_FREEZABLE | WQ_PERCPU, 0);
>   	if (!mdp->job_wq) {
>   		dev_err(dev, "Unable to create job workqueue\n");
>   		ret = -ENOMEM;
>   		goto err_deinit_comp;
>   	}
>   
> -	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock", WQ_FREEZABLE,
> +	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock",
> +					WQ_FREEZABLE | WQ_PERCPU,
>   					0);
>   	if (!mdp->clock_wq) {
>   		dev_err(dev, "Unable to create clock workqueue\n");

Hi,

Le vendredi 07 novembre 2025 à 15:13 +0100, Marco Crivellari a écrit :
> Currently if a user enqueues a work item using schedule_delayed_work() the
> used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
> WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
> schedule_work() that is using system_wq and queue_work(), that makes use
> again of WORK_CPU_UNBOUND.
> This lack of consistency cannot be addressed without refactoring the API.
> 
> alloc_workqueue() treats all queues as per-CPU by default, while unbound
> workqueues must opt-in via WQ_UNBOUND.
> 
> This default is suboptimal: most workloads benefit from unbound queues,
> allowing the scheduler to place worker threads where they’re needed and
> reducing noise when CPUs are isolated.
> 
> This continues the effort to refactor workqueue APIs, which began with
> the introduction of new workqueues and a new alloc_workqueue flag in:
> 
> commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
> commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
> 
> This change adds a new WQ_PERCPU flag to explicitly request
> alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
> 
> With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
> any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
> must now use WQ_PERCPU.
> 
> Once migration is complete, WQ_UNBOUND can be removed and unbound will
> become the implicit default.

I have to admit, there is likely no review here due to the lack of knowledge, so
in order to help educate myself (hopefully its not just me), can you explain why
the new default of WQ_UNBOUND would not be a fit for this driver ? After all,
the author didn't care and didn't make a choice, so I feel like its worth
asking.

cheers,
Nicolas

> 
> Suggested-by: Tejun Heo <tj@kernel.org>
> Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
> ---
>  drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> index 6559d72d5d42..9083367ae2e4 100644
> --- a/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> +++ b/drivers/media/platform/mediatek/mdp3/mtk-mdp3-core.c
> @@ -255,14 +255,16 @@ static int mdp_probe(struct platform_device *pdev)
>  		goto err_free_mutex;
>  	}
>  
> -	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME, WQ_FREEZABLE, 0);
> +	mdp->job_wq = alloc_workqueue(MDP_MODULE_NAME,
> +				      WQ_FREEZABLE | WQ_PERCPU, 0);
>  	if (!mdp->job_wq) {
>  		dev_err(dev, "Unable to create job workqueue\n");
>  		ret = -ENOMEM;
>  		goto err_deinit_comp;
>  	}
>  
> -	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock",
> WQ_FREEZABLE,
> +	mdp->clock_wq = alloc_workqueue(MDP_MODULE_NAME "-clock",
> +					WQ_FREEZABLE | WQ_PERCPU,
>  					0);
>  	if (!mdp->clock_wq) {
>  		dev_err(dev, "Unable to create clock workqueue\n");

On Tue, Dec 9, 2025 at 9:57 PM Nicolas Dufresne <nicolas@ndufresne.ca> wrote:
>
> Hi,
> I have to admit, there is likely no review here due to the lack of knowledge, so
> in order to help educate myself (hopefully its not just me), can you explain why
> the new default of WQ_UNBOUND would not be a fit for this driver ? After all,
> the author didn't care and didn't make a choice, so I feel like its worth
> asking.

Hi Nicolas,

The fact is that "alloc_workqueue()" without WQ_UNBOUND it means per-cpu.
So what we are doing here is just make explicit that the workqueue is per-cpu.

Currently there are no behavioral changes in alloc_workqueue(); in a
future release
WQ_UNBOUND will be removed and unbound will be the default. But as for now,
it is still per-cpu.

We can of course change the current behavior and I can send the v2 with
WQ_UNBOUND instead. Looking at the code there are not per-cpu variable and
the workqueue does not have the WQ_BH flag, so we can convert it if it
is better.

Thanks!

--

Marco Crivellari

L3 Support Engineer

Hi,

Le mercredi 10 décembre 2025 à 16:30 +0100, Marco Crivellari a écrit :
> On Tue, Dec 9, 2025 at 9:57 PM Nicolas Dufresne <nicolas@ndufresne.ca> wrote:
> > 
> > Hi,
> > I have to admit, there is likely no review here due to the lack of knowledge, so
> > in order to help educate myself (hopefully its not just me), can you explain why
> > the new default of WQ_UNBOUND would not be a fit for this driver ? After all,
> > the author didn't care and didn't make a choice, so I feel like its worth
> > asking.
> 
> Hi Nicolas,
> 
> The fact is that "alloc_workqueue()" without WQ_UNBOUND it means per-cpu.
> So what we are doing here is just make explicit that the workqueue is per-cpu.
> 
> Currently there are no behavioral changes in alloc_workqueue(); in a
> future release
> WQ_UNBOUND will be removed and unbound will be the default. But as for now,
> it is still per-cpu.
> 
> We can of course change the current behavior and I can send the v2 with
> WQ_UNBOUND instead. Looking at the code there are not per-cpu variable and
> the workqueue does not have the WQ_BH flag, so we can convert it if it
> is better.

thanks for clarifying. This driver having no clear maintainer, it is hard to
delegate the checks needed, but from the description, it pretty much sounded as
if most driver are picking up the wrong thing, because that is what the default
do.

I don't have strong opinion, if you think this driver can be ported in one step,
that is always my preference, and making things explicit is also nice. But I'm
also fine picking this as-is for now. Let me know, your preference, available
time and safety of not breaking anything is valid argument to me.

Nicolas

Hi,

On Wed, Dec 10, 2025 at 5:04 PM Nicolas Dufresne <nicolas@ndufresne.ca> wrote:
> I don't have strong opinion, if you think this driver can be ported in one step,
> that is always my preference, and making things explicit is also nice. But I'm
> also fine picking this as-is for now. Let me know, your preference, available
> time and safety of not breaking anything is valid argument to me.

I would like to wait before the conversion in this case, to avoid
breaking the driver.
So I would like to keep WQ_PERCPU instead, like it is now.

Many thanks, and sorry for the late reply.

-- 

Marco Crivellari

L3 Support Engineer