[PATCH 0/2] driver core: remove software node from platform devices on device release

[PATCH 0/2] driver core: remove software node from platform devices on device release

[Bartosz Golaszewski]

This fixes an issue in platform device code where, if we specify a
software node for a platform device using struct platform_device_info,
it will not be removed on device .release().

The second patch adds a test-case that can be used to reproduce the
problem and prove that the fix works.

First patch should go into v7.1 while patch 2/2 can be queued for v7.2.

Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
---
Bartosz Golaszewski (2):
      driver core: platform: remove software node on release()
      driver core: platform: tests: add a test case for correct swnode removal

 drivers/base/platform.c                  |  1 +
 drivers/base/test/platform-device-test.c | 62 ++++++++++++++++++++++++++++++++
 2 files changed, 63 insertions(+)
---
base-commit: 3fa7d958829eb9bc3b469ed07f11de3d2804ef71
change-id: 20260410-swnode-remove-on-dev-unreg-42bfc4b23ba8

Best regards,
-- 
Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>

[PATCH 1/2] driver core: platform: remove software node on release()

[Bartosz Golaszewski]

If we pass a software node to a newly created device using struct
platform_device_info, it will not be removed when the device is
released. This may happen when a module creating the device is removed
or on failure in platform_device_add().

When we try to reuse that software node in a subsequent call to
platform_device_register_full(), it will fails with -EBUSY. Add the
missing call to device_remove_software_node() in release path.

Fixes: 0fc434bc2c45 ("driver core: platform: allow attaching software nodes when creating devices")
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
---
 drivers/base/platform.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 75b4698d0e582e67adafa78c312d75c72fd654cf..a442abb4f1028a91f3c56b51211335678f7f18a2 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -600,6 +600,7 @@ static void platform_device_release(struct device *dev)
 						  pdev.dev);
 
 	of_node_put(pa->pdev.dev.of_node);
+	device_remove_software_node(dev);
 	kfree(pa->pdev.dev.platform_data);
 	kfree(pa->pdev.mfd_cell);
 	kfree(pa->pdev.resource);

-- 
2.47.3

[PATCH 2/2] driver core: platform: tests: add a test case for correct swnode removal

[Bartosz Golaszewski]

Extend the kunit module for platform devices with a test case verifying
that the same software node can be added to platform devices repeatedly.

Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
---
 drivers/base/test/platform-device-test.c | 62 ++++++++++++++++++++++++++++++++
 1 file changed, 62 insertions(+)

diff --git a/drivers/base/test/platform-device-test.c b/drivers/base/test/platform-device-test.c
index 6355a2231b741791b54eb78af42e13f31f745184..9046be1cd0f68b57f70eddb15227609c925318ce 100644
--- a/drivers/base/test/platform-device-test.c
+++ b/drivers/base/test/platform-device-test.c
@@ -5,8 +5,10 @@
 
 #include <linux/device.h>
 #include <linux/device/bus.h>
+#include <linux/fwnode.h>
 #include <linux/of_platform.h>
 #include <linux/platform_device.h>
+#include <linux/property.h>
 
 #define DEVICE_NAME "test"
 
@@ -253,9 +255,69 @@ static struct kunit_suite platform_device_match_test_suite = {
 	.test_cases = platform_device_match_tests,
 };
 
+static int platform_device_swnode_test_probe(struct platform_device *pdev)
+{
+	return 0;
+}
+
+static struct platform_driver platform_swnode_test_driver = {
+	.probe = platform_device_swnode_test_probe,
+	.driver = {
+		.name = DEVICE_NAME,
+	},
+};
+
+static const struct software_node platform_device_test_swnode = { };
+
+static void platform_device_swnode_add_twice(struct kunit *test)
+{
+	struct platform_device_info pdevinfo;
+	struct platform_device *pdev;
+	struct fwnode_handle fwnode;
+	int ret;
+
+	ret = kunit_platform_driver_register(test, &platform_swnode_test_driver);
+	KUNIT_ASSERT_EQ(test, ret, 0);
+
+	fwnode_init(&fwnode, NULL);
+	pdevinfo = (struct platform_device_info){
+		.name = DEVICE_NAME,
+		.id = PLATFORM_DEVID_NONE,
+		.fwnode = &fwnode,
+		.swnode = &platform_device_test_swnode,
+	};
+
+	pdev = platform_device_register_full(&pdevinfo);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, pdev);
+
+	wait_for_device_probe();
+	KUNIT_ASSERT_TRUE(test, device_is_bound(&pdev->dev));
+
+	platform_device_unregister(pdev);
+
+	pdev = platform_device_register_full(&pdevinfo);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, pdev);
+
+	wait_for_device_probe();
+	KUNIT_ASSERT_TRUE(test, device_is_bound(&pdev->dev));
+
+	platform_device_unregister(pdev);
+}
+
+static struct kunit_case platform_device_swnode_tests[] = {
+	KUNIT_CASE(platform_device_swnode_add_twice),
+	{}
+};
+
+static struct kunit_suite platform_device_swnode_test_suite = {
+	.name = "platform-device-swnode",
+	.test_cases = platform_device_swnode_tests,
+};
+
 kunit_test_suites(
 	&platform_device_devm_test_suite,
 	&platform_device_match_test_suite,
+	&platform_device_swnode_test_suite,
 );
 
 MODULE_DESCRIPTION("Test module for platform devices");

-- 
2.47.3

Re: [PATCH 1/2] driver core: platform: remove software node on release()

[Dmitry Torokhov]

Hi Bartosz,

On Fri, Apr 10, 2026 at 01:50:45PM +0200, Bartosz Golaszewski wrote:
> If we pass a software node to a newly created device using struct
> platform_device_info, it will not be removed when the device is
> released. This may happen when a module creating the device is removed
> or on failure in platform_device_add().
> 
> When we try to reuse that software node in a subsequent call to
> platform_device_register_full(), it will fails with -EBUSY. Add the
> missing call to device_remove_software_node() in release path.

Thank you for spotting this, but I do not believe the patch is correct
as it stands. We need to differentiate between nodes registered by
platform devices vs nodes that already registered where we simply use
them.

I think we need to mark nodes that platform_device_register_full()
registers form swnodes as "managed = true" so that they are cleaned up
properly.

See also sashiko feedback.

https://sashiko.dev/#/patchset/20260410-swnode-remove-on-dev-unreg-v1-0-cd7d305f3db2%40oss.qualcomm.com

Thanks.

-- 
Dmitry

Re: [PATCH 1/2] driver core: platform: remove software node on release()

[Bartosz Golaszewski]

On Mon, Apr 13, 2026 at 5:56 AM Dmitry Torokhov
<dmitry.torokhov@gmail.com> wrote:
>
> Hi Bartosz,
>
> On Fri, Apr 10, 2026 at 01:50:45PM +0200, Bartosz Golaszewski wrote:
> > If we pass a software node to a newly created device using struct
> > platform_device_info, it will not be removed when the device is
> > released. This may happen when a module creating the device is removed
> > or on failure in platform_device_add().
> >
> > When we try to reuse that software node in a subsequent call to
> > platform_device_register_full(), it will fails with -EBUSY. Add the
> > missing call to device_remove_software_node() in release path.
>
> Thank you for spotting this, but I do not believe the patch is correct
> as it stands. We need to differentiate between nodes registered by
> platform devices vs nodes that already registered where we simply use
> them.
>

We already do: platform_device_release() is only set for devices
dynamically allocated with platform_device_alloc() which is what
happens in platform_device_register_full(). platform_device_alloc()
already imposes limitations on the caller - like having to provide
platform data that can be shallow-copied as platform_device_release()
calls kfree() on it unconditionally. Even if the user adds the
software node manually, it needs to be removed.

> I think we need to mark nodes that platform_device_register_full()
> registers form swnodes as "managed = true" so that they are cleaned up
> properly.
>
> See also sashiko feedback.
>
> https://sashiko.dev/#/patchset/20260410-swnode-remove-on-dev-unreg-v1-0-cd7d305f3db2%40oss.qualcomm.com
>

I'll see about reproducing it and possibly adding a kunit test-case as well.

Bart