[PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[Marco Crivellari]

The workqueue API introduced a new flag, WQ_PERCPU, that has to be used when
WQ_UNBOUND is not present. One of these flags must be present, but not
both of them.

To limit usage mistakes, emit an ERROR if one of the below condition is met:
- alloc_workqueue() is called without WQ_PERCPU nor WQ_UNBOUND
- alloc_workqueue() is called with both WQ_PERCPU and WQ_UNBOUND

Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
---
 scripts/checkpatch.pl | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 0492d6afc9a1..2706aba6dac0 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -7804,6 +7804,22 @@ sub process {
 			ERROR("UNINITIALIZED_PTR_WITH_FREE",
 			      "pointer '$1' with __free attribute should be initialized\n" . $herecurr);
 		}
+
+# check alloc_workqueue() flags parameter (2nd argument) for WQ_PERCPU, WQ_UNBOUND presence
+		if ($line =~ /\balloc_workqueue\s*\(/ &&
+			$stat =~ /\balloc_workqueue\s*\(\s*[^,]+\s*,\s*([^,]+)/s) {
+			my $flags = $1;
+			my $has_percpu = $flags =~ /\bWQ_PERCPU\b/;
+			my $has_unbound = $flags =~ /\bWQ_UNBOUND\b/;
+
+			if ($has_percpu && $has_unbound) {
+				ERROR("ALLOC_WORKQUEUE_FLAGS",
+				      "alloc_workqueue() second parameter should not contain both WQ_PERCPU and WQ_UNBOUND\n" . $herecurr);
+			} elsif (!$has_percpu && !$has_unbound) {
+				ERROR("ALLOC_WORKQUEUE_FLAGS",
+				      "alloc_workqueue() second parameter should specify either WQ_PERCPU or WQ_UNBOUND\n" . $herecurr);
+			}
+		}
 	}
 
 	# If we have no input at all, then there is nothing to report on
-- 
2.54.0

Re: [PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[Joe Perches]

On Wed, 2026-06-03 at 16:09 +0200, Marco Crivellari wrote:
> The workqueue API introduced a new flag, WQ_PERCPU, that has to be used when
> WQ_UNBOUND is not present. One of these flags must be present, but not
> both of them.
> 
> To limit usage mistakes, emit an ERROR if one of the below condition is met:
> - alloc_workqueue() is called without WQ_PERCPU nor WQ_UNBOUND
> - alloc_workqueue() is called with both WQ_PERCPU and WQ_UNBOUND
[]
> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
[]
> @@ -7804,6 +7804,22 @@ sub process {
>  			ERROR("UNINITIALIZED_PTR_WITH_FREE",
>  			      "pointer '$1' with __free attribute should be initialized\n" . $herecurr);
>  		}
> +
> +# check alloc_workqueue() flags parameter (2nd argument) for WQ_PERCPU, WQ_UNBOUND presence
> +		if ($line =~ /\balloc_workqueue\s*\(/ &&
> +			$stat =~ /\balloc_workqueue\s*\(\s*[^,]+\s*,\s*([^,]+)/s) {
> +			my $flags = $1;
> +			my $has_percpu = $flags =~ /\bWQ_PERCPU\b/;
> +			my $has_unbound = $flags =~ /\bWQ_UNBOUND\b/;
> +
> +			if ($has_percpu && $has_unbound) {
> +				ERROR("ALLOC_WORKQUEUE_FLAGS",
> +				      "alloc_workqueue() second parameter should not contain both WQ_PERCPU and WQ_UNBOUND\n" . $herecurr);
> +			} elsif (!$has_percpu && !$has_unbound) {
> +				ERROR("ALLOC_WORKQUEUE_FLAGS",
> +				      "alloc_workqueue() second parameter should specify either WQ_PERCPU or WQ_UNBOUND\n" . $herecurr);
> +			}
> +		}
>  	}

OK but a couple comments:

Add devm_alloc_workqueue as well?

Avoid using $line here as $stat isn't populated on multi-line uses
of alloc_workqueue.  Just test $stat

Using the patch below gives a few matches treewide with a few
false positives as the flags are stored in a variable in
drivers/nvme/host/... and fs/btrfs/...

-----------------
block/blk-zoned.c
-----------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1926: FILE: block/blk-zoned.c:1926:
+	disk->zone_wplugs_wq =
+		alloc_workqueue("%s_zwplugs", WQ_MEM_RECLAIM | WQ_HIGHPRI,
+				pool_size, disk->disk_name);

-----------------------------------------
drivers/gpu/drm/bridge/analogix/anx7625.c
-----------------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#2851: FILE: drivers/gpu/drm/bridge/analogix/anx7625.c:2851:
+		platform->workqueue = alloc_workqueue("anx7625_work",
+						      WQ_FREEZABLE | WQ_MEM_RECLAIM, 1);

-------------------------------------
drivers/gpu/drm/nouveau/nouveau_drm.c
-------------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#635: FILE: drivers/gpu/drm/nouveau/nouveau_drm.c:635:
+	drm->sched_wq = alloc_workqueue("nouveau_sched_wq_shared", 0,
+					WQ_MAX_ACTIVE);

---------------------------------------
drivers/gpu/drm/nouveau/nouveau_sched.c
---------------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#417: FILE: drivers/gpu/drm/nouveau/nouveau_sched.c:417:
+		wq = alloc_workqueue("nouveau_sched_wq_%d", 0, WQ_MAX_ACTIVE,
+				     current->pid);

------------------------------------------
drivers/media/pci/ddbridge/ddbridge-core.c
------------------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#3433: FILE: drivers/media/pci/ddbridge/ddbridge-core.c:3433:
+	ddb_wq = alloc_workqueue("ddbridge", 0, 0);

-------------------------------------
drivers/net/wireless/ath/ath6kl/usb.c
-------------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#639: FILE: drivers/net/wireless/ath/ath6kl/usb.c:639:
+	ar_usb->wq = alloc_workqueue("ath6kl_wq", 0, 0);

------------------------
drivers/nvme/host/core.c
------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#5410: FILE: drivers/nvme/host/core.c:5410:
+	nvme_wq = alloc_workqueue("nvme-wq", wq_flags, 0);

ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#5414: FILE: drivers/nvme/host/core.c:5414:
+	nvme_reset_wq = alloc_workqueue("nvme-reset-wq", wq_flags, 0);

ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#5418: FILE: drivers/nvme/host/core.c:5418:
+	nvme_delete_wq = alloc_workqueue("nvme-delete-wq", wq_flags, 0);

---------------------
drivers/rapidio/rio.c
---------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1997: FILE: drivers/rapidio/rio.c:1997:
+	rio_wq = alloc_workqueue("riodisc", 0, 0);

------------------------------
drivers/thermal/thermal_core.c
------------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1892: FILE: drivers/thermal/thermal_core.c:1892:
+	thermal_wq = alloc_workqueue("thermal_events", WQ_POWER_EFFICIENT, 0);

-------------------------
drivers/virt/acrn/irqfd.c
-------------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#209: FILE: drivers/virt/acrn/irqfd.c:209:
+	vm->irqfd_wq = alloc_workqueue("acrn_irqfd-%u", 0, 0, vm->vmid);

------------------
fs/btrfs/disk-io.c
------------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1950: FILE: fs/btrfs/disk-io.c:1950:
+	fs_info->endio_workers =
+		alloc_workqueue("btrfs-endio", flags, max_active);

ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1952: FILE: fs/btrfs/disk-io.c:1952:
+	fs_info->endio_meta_workers =
+		alloc_workqueue("btrfs-endio-meta", flags, max_active);

ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#1954: FILE: fs/btrfs/disk-io.c:1954:
+	fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);

----------------
fs/btrfs/scrub.c
----------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#3050: FILE: fs/btrfs/scrub.c:3050:
+	scrub_workers = alloc_workqueue("btrfs-scrub", flags, max_active);

---------------
fs/ntfs/super.c
---------------
ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
#2652: FILE: fs/ntfs/super.c:2652:
+	ntfs_wq = alloc_workqueue("ntfs-bg-io", 0, 0);

---
 scripts/checkpatch.pl | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 0492d6afc9a1..aed83ce4b552 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -7804,6 +7804,31 @@ sub process {
 			ERROR("UNINITIALIZED_PTR_WITH_FREE",
 			      "pointer '$1' with __free attribute should be initialized\n" . $herecurr);
 		}
+
+# check alloc_workqueue() and devm_alloc_workqueue parameters for
+# WQ_PERCPU and WQ_UNBOUND uses
+		if (defined($stat) &&
+		    $stat =~ /^[ \+]\s*(?:$Lval\s*=\s*)?((?:devm_)?alloc_workqueue)\s*($balanced_parens)/) {
+			my $func = $1;
+			my $args = $2;
+			my $has_percpu = $args =~ /\bWQ_PERCPU\b/;
+			my $has_unbound = $args =~ /\bWQ_UNBOUND\b/;
+			my $error_msg;
+
+			if ($has_percpu && $has_unbound) {
+				$error_msg = "$func() should not contain both WQ_PERCPU and WQ_UNBOUND\n";
+			} elsif (!$has_percpu && !$has_unbound) {
+				$error_msg = "$func() must specify either WQ_PERCPU or WQ_UNBOUND\n";
+			}
+
+			if (defined($error_msg)) {
+				my $stmt_cnt = statement_rawlines($stat);
+				my $herectx = get_stat_here($linenr, $stmt_cnt, $here);
+				ERROR("ALLOC_WORKQUEUE_FLAGS",
+				      $error_msg . $herectx);
+			}
+
+		}
 	}
 
 	# If we have no input at all, then there is nothing to report on

Re: [PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[Marco Crivellari]

Hi,

On Thu, Jun 4, 2026 at 7:08 AM Joe Perches <joe@perches.com> wrote:
> [...]
> OK but a couple comments:
>
> Add devm_alloc_workqueue as well?

No I don't think it would be useful / needed. There is only 1 user of
that macro, devm_alloc_ordered_workqueue(), which already specifies
the WQ_UNBOUND flag and it is part of the workqueue code.

Also, we're adding this in checkpatch, because it reflects what is
going to be present in the workqueue code:

https://lore.kernel.org/all/20260529130640.220328-3-marco.crivellari@suse.com/

> Avoid using $line here as $stat isn't populated on multi-line uses
> of alloc_workqueue.  Just test $stat

Aha, noted. I will remove that test.

> Using the patch below gives a few matches treewide with a few
> false positives as the flags are stored in a variable in
> drivers/nvme/host/... and fs/btrfs/...

Yeah, those false positives were expected. I thought about adding a
message about that, indeed.

> -----------------
> block/blk-zoned.c
> -----------------
> ERROR: alloc_workqueue() must specify either WQ_PERCPU or WQ_UNBOUND
> #1926: FILE: block/blk-zoned.c:1926:
> +       disk->zone_wplugs_wq =
> +               alloc_workqueue("%s_zwplugs", WQ_MEM_RECLAIM | WQ_HIGHPRI,
> +                               pool_size, disk->disk_name);

Interestingly, my Coccinelle script didn't match this.

>  scripts/checkpatch.pl | 25 +++++++++++++++++++++++++
>  1 file changed, 25 insertions(+)
>
> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> index 0492d6afc9a1..aed83ce4b552 100755
> --- a/scripts/checkpatch.pl
> +++ b/scripts/checkpatch.pl
> @@ -7804,6 +7804,31 @@ sub process {
>                         ERROR("UNINITIALIZED_PTR_WITH_FREE",
>                               "pointer '$1' with __free attribute should be initialized\n" . $herecurr);
>                 }
> +
> +# check alloc_workqueue() and devm_alloc_workqueue parameters for
> +# WQ_PERCPU and WQ_UNBOUND uses
> +               if (defined($stat) &&
> +                   $stat =~ /^[ \+]\s*(?:$Lval\s*=\s*)?((?:devm_)?alloc_workqueue)\s*($balanced_parens)/) {
> +                       my $func = $1;
> +                       my $args = $2;
> +                       my $has_percpu = $args =~ /\bWQ_PERCPU\b/;
> +                       my $has_unbound = $args =~ /\bWQ_UNBOUND\b/;
> +                       my $error_msg;
> +
> +                       if ($has_percpu && $has_unbound) {
> +                               $error_msg = "$func() should not contain both WQ_PERCPU and WQ_UNBOUND\n";
> +                       } elsif (!$has_percpu && !$has_unbound) {
> +                               $error_msg = "$func() must specify either WQ_PERCPU or WQ_UNBOUND\n";
> +                       }
> +
> +                       if (defined($error_msg)) {
> +                               my $stmt_cnt = statement_rawlines($stat);
> +                               my $herectx = get_stat_here($linenr, $stmt_cnt, $here);
> +                               ERROR("ALLOC_WORKQUEUE_FLAGS",
> +                                     $error_msg . $herectx);
> +                       }
> +
> +               }
>         }
>
>         # If we have no input at all, then there is nothing to report on

I will integrate my code with this!

Thanks!

-- 

Marco Crivellari

SUSE Labs

Re: [PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[Joe Perches]

On Thu, 2026-06-04 at 11:31 +0200, Marco Crivellari wrote:
> On Thu, Jun 4, 2026 at 7:08 AM Joe Perches <[joe@perches.com](mailto:joe@perches.com)> wrote:
> [...]
> > OK but a couple comments:
> > 
> > Add devm_alloc_workqueue as well?
> 
> No I don't think it would be useful / needed. There is only 1 user of
> that macro, devm_alloc_ordered_workqueue(), which already specifies
> the WQ_UNBOUND flag and it is part of the workqueue code.

drivers/power/supply/ipaq_micro_battery.c:      mb->wq = devm_alloc_workqueue(&pdev->dev, "ipaq-battery-wq",

Maybe there will be more.

Re: [PATCH] checkpatch: Check WQ_PERCPU or WQ_UNBOUND presence in alloc_workqueue() users

[Marco Crivellari]

On Thu, Jun 4, 2026 at 4:09 PM Joe Perches <joe@perches.com> wrote:
>
> On Thu, 2026-06-04 at 11:31 +0200, Marco Crivellari wrote:
> > On Thu, Jun 4, 2026 at 7:08 AM Joe Perches <[joe@perches.com](mailto:joe@perches.com)> wrote:
> > [...]
> > > OK but a couple comments:
> > >
> > > Add devm_alloc_workqueue as well?
> >
> > No I don't think it would be useful / needed. There is only 1 user of
> > that macro, devm_alloc_ordered_workqueue(), which already specifies
> > the WQ_UNBOUND flag and it is part of the workqueue code.
>
> drivers/power/supply/ipaq_micro_battery.c:      mb->wq = devm_alloc_workqueue(&pdev->dev, "ipaq-battery-wq",
>
> Maybe there will be more.

Yes it is the only one in v7.1-rc6.

BTW, let's keep the new regex for both macros, it's not a problem, so
we can cover both cases.

Thanks!

-- 

Marco Crivellari

SUSE Labs