Linux driver-core infrastructure
 help / color / mirror / Atom feed
* [PATCH] software node: make swnodes hold their own references on parents
@ 2026-11-06 11:13 Bartosz Golaszewski
  2026-11-09 19:52 ` Andy Shevchenko
  2026-11-17 12:10 ` Johan Hovold
  0 siblings, 2 replies; 6+ messages in thread
From: Bartosz Golaszewski @ 2026-11-06 11:13 UTC (permalink / raw)
  To: Andy Shevchenko, Daniel Scally, Heikki Krogerus, Sakari Ailus,
	Bartosz Golaszewski, Greg Kroah-Hartman, Rafael J. Wysocki,
	Danilo Krummrich, Johan Hovold, Linus Walleij
  Cc: Rafael J. Wysocki, linux-acpi, driver-core, linux-kernel, stable,
	Bartosz Golaszewski

In swnode_register(), swnode->parent is assigned before calling
kobject_init_and_add(), which normally acquires a reference on
parent->kobj via kobject_add_internal() that is held until after
software_node_release() completes. However, if kobject_init_and_add()
fails before acquiring the parent reference or drops it on error in
kobject_add_internal(), swnode->kobj holds no reference on parent->kobj
while swnode->parent remains set.

With CONFIG_DEBUG_KOBJECT_RELEASE enabled, the random delay in
kobject_delayed_cleanup() may create a situation where - when
unregistering the child node and then the parent - the parent object
gets released before the child who's left with a dangling pointer.

Acquire a refernce to the parent swnode when registering the child and
drop it on child's release.

Cc: stable@vger.kernel.org
Fixes: 59abd83672f7 ("drivers: base: Introducing software nodes to the firmware node framework")
Closes: https://sashiko.dev/#/bug/linux-82933c14-d212-49f1-bd2f-d4ab1e16fed7
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
---
 drivers/base/swnode.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
index 84a77ffa3dc4bbcca62bbe67c3390693500b01ea..986dcb933d2c9274e9000968f115d7fccc2d103f 100644
--- a/drivers/base/swnode.c
+++ b/drivers/base/swnode.c
@@ -843,11 +843,12 @@ static void software_node_free(const struct software_node *node)
 static void software_node_release(struct kobject *kobj)
 {
 	struct swnode *swnode = kobj_to_swnode(kobj);
+	struct swnode *parent = swnode->parent;
 
 	fwnode_links_purge(&swnode->fwnode);
 
-	if (swnode->parent) {
-		ida_free(&swnode->parent->child_ids, swnode->id);
+	if (parent) {
+		ida_free(&parent->child_ids, swnode->id);
 		list_del(&swnode->entry);
 	} else {
 		ida_free(&swnode_root_ids, swnode->id);
@@ -858,6 +859,9 @@ static void software_node_release(struct kobject *kobj)
 
 	ida_destroy(&swnode->child_ids);
 	kfree(swnode);
+
+	if (parent)
+		swnode_put(parent);
 }
 
 static const struct kobj_type software_node_type = {
@@ -885,7 +889,7 @@ swnode_register(const struct software_node *node, struct swnode *parent,
 
 	swnode->id = ret;
 	swnode->node = node;
-	swnode->parent = parent;
+	swnode->parent = parent ? swnode_get(parent) : NULL;
 	swnode->kobj.kset = swnode_kset;
 	fwnode_init(&swnode->fwnode, &software_node_ops);
 

---
base-commit: ddfe51b0a0f5dc9e415cc2c8ad8db17532dd426b
change-id: 20261106-swnode-kobject-parent-ref-3bd9b9b75aeb

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


^ permalink raw reply related	[flat|nested] 6+ messages in thread

* Re: [PATCH] software node: make swnodes hold their own references on parents
  2026-11-06 11:13 [PATCH] software node: make swnodes hold their own references on parents Bartosz Golaszewski
@ 2026-11-09 19:52 ` Andy Shevchenko
  2026-11-10  8:57   ` Bartosz Golaszewski
  2026-11-17 12:10 ` Johan Hovold
  1 sibling, 1 reply; 6+ messages in thread
From: Andy Shevchenko @ 2026-11-09 19:52 UTC (permalink / raw)
  To: Bartosz Golaszewski
  Cc: Daniel Scally, Heikki Krogerus, Sakari Ailus, Bartosz Golaszewski,
	Greg Kroah-Hartman, Rafael J. Wysocki, Danilo Krummrich,
	Johan Hovold, Linus Walleij, Rafael J. Wysocki, linux-acpi,
	driver-core, linux-kernel, stable

On Fri, Nov 06, 2026 at 12:13:28PM +0100, Bartosz Golaszewski wrote:
> In swnode_register(), swnode->parent is assigned before calling
> kobject_init_and_add(), which normally acquires a reference on
> parent->kobj via kobject_add_internal() that is held until after
> software_node_release() completes. However, if kobject_init_and_add()
> fails before acquiring the parent reference or drops it on error in
> kobject_add_internal(), swnode->kobj holds no reference on parent->kobj
> while swnode->parent remains set.
> 
> With CONFIG_DEBUG_KOBJECT_RELEASE enabled, the random delay in
> kobject_delayed_cleanup() may create a situation where - when
> unregistering the child node and then the parent - the parent object
> gets released before the child who's left with a dangling pointer.
> 
> Acquire a refernce to the parent swnode when registering the child and
> drop it on child's release.

...

> +	if (parent)
> +		swnode_put(parent);

This will be the second case of potentially NULL being passed to swnode_put().
Since it's a static internal API, I would just add a check there and hope that
compiler will inline and that will be optimized away on the cases when the
parameter is known not to be NULL.

Maybe in a separate change.

...

> -	swnode->parent = parent;
> +	swnode->parent = parent ? swnode_get(parent) : NULL;

For the consistency's sake, same to swnode_get().

Note, that it also has a case in the current code. So with that patch it will
be two users of each.

-- 
With Best Regards,
Andy Shevchenko



^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] software node: make swnodes hold their own references on parents
  2026-11-09 19:52 ` Andy Shevchenko
@ 2026-11-10  8:57   ` Bartosz Golaszewski
  0 siblings, 0 replies; 6+ messages in thread
From: Bartosz Golaszewski @ 2026-11-10  8:57 UTC (permalink / raw)
  To: Andy Shevchenko
  Cc: Bartosz Golaszewski, Daniel Scally, Heikki Krogerus, Sakari Ailus,
	Greg Kroah-Hartman, Rafael J. Wysocki, Danilo Krummrich,
	Johan Hovold, Linus Walleij, Rafael J. Wysocki, linux-acpi,
	driver-core, linux-kernel, stable

On Mon, Nov 09, 2026 at 09:52:12PM +0200, Andy Shevchenko wrote:
> On Fri, Nov 06, 2026 at 12:13:28PM +0100, Bartosz Golaszewski wrote:
> > In swnode_register(), swnode->parent is assigned before calling
> > kobject_init_and_add(), which normally acquires a reference on
> > parent->kobj via kobject_add_internal() that is held until after
> > software_node_release() completes. However, if kobject_init_and_add()
> > fails before acquiring the parent reference or drops it on error in
> > kobject_add_internal(), swnode->kobj holds no reference on parent->kobj
> > while swnode->parent remains set.
> > 
> > With CONFIG_DEBUG_KOBJECT_RELEASE enabled, the random delay in
> > kobject_delayed_cleanup() may create a situation where - when
> > unregistering the child node and then the parent - the parent object
> > gets released before the child who's left with a dangling pointer.
> > 
> > Acquire a refernce to the parent swnode when registering the child and
> > drop it on child's release.
> 
> ...
> 
> > +	if (parent)
> > +		swnode_put(parent);
> 
> This will be the second case of potentially NULL being passed to swnode_put().
> Since it's a static internal API, I would just add a check there and hope that
> compiler will inline and that will be optimized away on the cases when the
> parameter is known not to be NULL.
> 
> Maybe in a separate change.
>

Yes, I actually added it to my TODO list but this is clearly a fix, so I
wanted to keep it small.

Bart

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] software node: make swnodes hold their own references on parents
  2026-11-06 11:13 [PATCH] software node: make swnodes hold their own references on parents Bartosz Golaszewski
  2026-11-09 19:52 ` Andy Shevchenko
@ 2026-11-17 12:10 ` Johan Hovold
  2026-11-17 13:39   ` Bartosz Golaszewski
  1 sibling, 1 reply; 6+ messages in thread
From: Johan Hovold @ 2026-11-17 12:10 UTC (permalink / raw)
  To: Bartosz Golaszewski
  Cc: Andy Shevchenko, Daniel Scally, Heikki Krogerus, Sakari Ailus,
	Bartosz Golaszewski, Greg Kroah-Hartman, Rafael J. Wysocki,
	Danilo Krummrich, Linus Walleij, Rafael J. Wysocki, linux-acpi,
	driver-core, linux-kernel, stable

On Fri, Nov 06, 2026 at 12:13:28PM +0100, Bartosz Golaszewski wrote:
> In swnode_register(), swnode->parent is assigned before calling
> kobject_init_and_add(), which normally acquires a reference on
> parent->kobj via kobject_add_internal() that is held until after
> software_node_release() completes. However, if kobject_init_and_add()
> fails before acquiring the parent reference or drops it on error in
> kobject_add_internal(), swnode->kobj holds no reference on parent->kobj
> while swnode->parent remains set.
> 
> With CONFIG_DEBUG_KOBJECT_RELEASE enabled, the random delay in
> kobject_delayed_cleanup() may create a situation where - when
> unregistering the child node and then the parent - the parent object
> gets released before the child who's left with a dangling pointer.
> 
> Acquire a refernce to the parent swnode when registering the child and
> drop it on child's release.
> 
> Cc: stable@vger.kernel.org
> Fixes: 59abd83672f7 ("drivers: base: Introducing software nodes to the firmware node framework")
> Closes: https://sashiko.dev/#/bug/linux-82933c14-d212-49f1-bd2f-d4ab1e16fed7
> Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
> ---
>  drivers/base/swnode.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
> index 84a77ffa3dc4bbcca62bbe67c3390693500b01ea..986dcb933d2c9274e9000968f115d7fccc2d103f 100644
> --- a/drivers/base/swnode.c
> +++ b/drivers/base/swnode.c
> @@ -843,11 +843,12 @@ static void software_node_free(const struct software_node *node)
>  static void software_node_release(struct kobject *kobj)
>  {
>  	struct swnode *swnode = kobj_to_swnode(kobj);
> +	struct swnode *parent = swnode->parent;
>  
>  	fwnode_links_purge(&swnode->fwnode);
>  
> -	if (swnode->parent) {
> -		ida_free(&swnode->parent->child_ids, swnode->id);
> +	if (parent) {
> +		ida_free(&parent->child_ids, swnode->id);
>  		list_del(&swnode->entry);

Why not simply

		swnode_put(swnode->parent);

here?

>  	} else {
>  		ida_free(&swnode_root_ids, swnode->id);
> @@ -858,6 +859,9 @@ static void software_node_release(struct kobject *kobj)
>  
>  	ida_destroy(&swnode->child_ids);
>  	kfree(swnode);
> +
> +	if (parent)
> +		swnode_put(parent);
>  }

Johan

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] software node: make swnodes hold their own references on parents
  2026-11-17 12:10 ` Johan Hovold
@ 2026-11-17 13:39   ` Bartosz Golaszewski
  2026-11-17 13:47     ` Johan Hovold
  0 siblings, 1 reply; 6+ messages in thread
From: Bartosz Golaszewski @ 2026-11-17 13:39 UTC (permalink / raw)
  To: Johan Hovold
  Cc: Bartosz Golaszewski, Andy Shevchenko, Daniel Scally,
	Heikki Krogerus, Sakari Ailus, Greg Kroah-Hartman,
	Rafael J. Wysocki, Danilo Krummrich, Linus Walleij,
	Rafael J. Wysocki, linux-acpi, driver-core, linux-kernel, stable

On Tue, Nov 17, 2026 at 01:10:02PM +0100, Johan Hovold wrote:
> On Fri, Nov 06, 2026 at 12:13:28PM +0100, Bartosz Golaszewski wrote:
> > In swnode_register(), swnode->parent is assigned before calling
> > kobject_init_and_add(), which normally acquires a reference on
> > parent->kobj via kobject_add_internal() that is held until after
> > software_node_release() completes. However, if kobject_init_and_add()
> > fails before acquiring the parent reference or drops it on error in
> > kobject_add_internal(), swnode->kobj holds no reference on parent->kobj
> > while swnode->parent remains set.
> > 
> > With CONFIG_DEBUG_KOBJECT_RELEASE enabled, the random delay in
> > kobject_delayed_cleanup() may create a situation where - when
> > unregistering the child node and then the parent - the parent object
> > gets released before the child who's left with a dangling pointer.
> > 
> > Acquire a refernce to the parent swnode when registering the child and
> > drop it on child's release.
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: 59abd83672f7 ("drivers: base: Introducing software nodes to the firmware node framework")
> > Closes: https://sashiko.dev/#/bug/linux-82933c14-d212-49f1-bd2f-d4ab1e16fed7
> > Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
> > ---
> >  drivers/base/swnode.c | 10 +++++++---
> >  1 file changed, 7 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
> > index 84a77ffa3dc4bbcca62bbe67c3390693500b01ea..986dcb933d2c9274e9000968f115d7fccc2d103f 100644
> > --- a/drivers/base/swnode.c
> > +++ b/drivers/base/swnode.c
> > @@ -843,11 +843,12 @@ static void software_node_free(const struct software_node *node)
> >  static void software_node_release(struct kobject *kobj)
> >  {
> >  	struct swnode *swnode = kobj_to_swnode(kobj);
> > +	struct swnode *parent = swnode->parent;
> >  
> >  	fwnode_links_purge(&swnode->fwnode);
> >  
> > -	if (swnode->parent) {
> > -		ida_free(&swnode->parent->child_ids, swnode->id);
> > +	if (parent) {
> > +		ida_free(&parent->child_ids, swnode->id);
> >  		list_del(&swnode->entry);
> 
> Why not simply
> 
> 		swnode_put(swnode->parent);
> 
> here?
>

Because swnode_put() doesn't accept NULL. I have it on my TODO list to
add it but this is a fix, so I want to keep it minimal.

Bart

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH] software node: make swnodes hold their own references on parents
  2026-11-17 13:39   ` Bartosz Golaszewski
@ 2026-11-17 13:47     ` Johan Hovold
  0 siblings, 0 replies; 6+ messages in thread
From: Johan Hovold @ 2026-11-17 13:47 UTC (permalink / raw)
  To: Bartosz Golaszewski
  Cc: Bartosz Golaszewski, Andy Shevchenko, Daniel Scally,
	Heikki Krogerus, Sakari Ailus, Greg Kroah-Hartman,
	Rafael J. Wysocki, Danilo Krummrich, Linus Walleij,
	Rafael J. Wysocki, linux-acpi, driver-core, linux-kernel, stable

On Tue, Nov 17, 2026 at 02:39:27PM +0100, Bartosz Golaszewski wrote:
> On Tue, Nov 17, 2026 at 01:10:02PM +0100, Johan Hovold wrote:
> > On Fri, Nov 06, 2026 at 12:13:28PM +0100, Bartosz Golaszewski wrote:

> > > @@ -843,11 +843,12 @@ static void software_node_free(const struct software_node *node)
> > >  static void software_node_release(struct kobject *kobj)
> > >  {
> > >  	struct swnode *swnode = kobj_to_swnode(kobj);
> > > +	struct swnode *parent = swnode->parent;
> > >  
> > >  	fwnode_links_purge(&swnode->fwnode);
> > >  
> > > -	if (swnode->parent) {
> > > -		ida_free(&swnode->parent->child_ids, swnode->id);
> > > +	if (parent) {
> > > +		ida_free(&parent->child_ids, swnode->id);
> > >  		list_del(&swnode->entry);
> > 
> > Why not simply
> > 
> > 		swnode_put(swnode->parent);
> > 
> > here?
> >
> 
> Because swnode_put() doesn't accept NULL. I have it on my TODO list to
> add it but this is a fix, so I want to keep it minimal.

swnode->parent is not NULL here in the conditional block.

Johan

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2026-11-17 13:47 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-06 11:13 [PATCH] software node: make swnodes hold their own references on parents Bartosz Golaszewski
2026-11-09 19:52 ` Andy Shevchenko
2026-11-10  8:57   ` Bartosz Golaszewski
2026-11-17 12:10 ` Johan Hovold
2026-11-17 13:39   ` Bartosz Golaszewski
2026-11-17 13:47     ` Johan Hovold

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox