[PATCH] of: Fix locking when calling of_get_next_available_child()

[PATCH] of: Fix locking when calling of_get_next_available_child()

[Grant Likely]

of_get_next_available_child() obtains the devtree_lock and then calls
of_device_is_available() which also attempts to claim the lock. This is
obviously incorrect and causes a deadlock on boot. Fix issue by adding
an variant of of_device_is_available() which doesn't obtain the lock.

Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Rob Herring <robherring2@gmail.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
---
 drivers/of/base.c |   15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/of/base.c b/drivers/of/base.c
index e8d65af..4af74b7 100644
--- a/drivers/of/base.c
+++ b/drivers/of/base.c
@@ -297,7 +297,7 @@ EXPORT_SYMBOL(of_machine_is_compatible);
  *  Returns 1 if the status property is absent or set to "okay" or "ok",
  *  0 otherwise
  */
-int of_device_is_available(const struct device_node *device)
+int __of_device_is_available(const struct device_node *device)
 {
 	const char *status;
 	int statlen;
@@ -313,6 +313,17 @@ int of_device_is_available(const struct device_node *device)
 
 	return 0;
 }
+
+int of_device_is_available(const struct device_node *device)
+{
+	unsigned long flags;
+	int rc;
+
+	raw_spin_lock_irqsave(&devtree_lock, flags);
+	rc = __of_device_is_available(device);
+	raw_spin_unlock_irqrestore(&devtree_lock, flags);
+	return rc;
+}
 EXPORT_SYMBOL(of_device_is_available);
 
 /**
@@ -404,7 +415,7 @@ struct device_node *of_get_next_available_child(const struct device_node *node,
 	raw_spin_lock(&devtree_lock);
 	next = prev ? prev->sibling : node->child;
 	for (; next; next = next->sibling) {
-		if (!of_device_is_available(next))
+		if (!__of_device_is_available(next))
 			continue;
 		if (of_node_get(next))
 			break;
-- 
1.7.10.4

[PATCH] of: Fix locking when calling of_get_next_available_child()

[Grant Likely]

of_get_next_available_child() obtains the devtree_lock and then calls
of_device_is_available() which also attempts to claim the lock. This is
obviously incorrect and causes a deadlock on boot. Fix issue by adding
an variant of of_device_is_available() which doesn't obtain the lock.

Cc: Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org>
Cc: Rob Herring <robherring2-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Signed-off-by: Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>
---
 drivers/of/base.c |   15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/of/base.c b/drivers/of/base.c
index e8d65af..4af74b7 100644
--- a/drivers/of/base.c
+++ b/drivers/of/base.c
@@ -297,7 +297,7 @@ EXPORT_SYMBOL(of_machine_is_compatible);
  *  Returns 1 if the status property is absent or set to "okay" or "ok",
  *  0 otherwise
  */
-int of_device_is_available(const struct device_node *device)
+int __of_device_is_available(const struct device_node *device)
 {
 	const char *status;
 	int statlen;
@@ -313,6 +313,17 @@ int of_device_is_available(const struct device_node *device)
 
 	return 0;
 }
+
+int of_device_is_available(const struct device_node *device)
+{
+	unsigned long flags;
+	int rc;
+
+	raw_spin_lock_irqsave(&devtree_lock, flags);
+	rc = __of_device_is_available(device);
+	raw_spin_unlock_irqrestore(&devtree_lock, flags);
+	return rc;
+}
 EXPORT_SYMBOL(of_device_is_available);
 
 /**
@@ -404,7 +415,7 @@ struct device_node *of_get_next_available_child(const struct device_node *node,
 	raw_spin_lock(&devtree_lock);
 	next = prev ? prev->sibling : node->child;
 	for (; next; next = next->sibling) {
-		if (!of_device_is_available(next))
+		if (!__of_device_is_available(next))
 			continue;
 		if (of_node_get(next))
 			break;
-- 
1.7.10.4

Re: [PATCH] of: Fix locking when calling of_get_next_available_child()

[Rob Herring]

On 02/11/2013 04:19 PM, Grant Likely wrote:
> of_get_next_available_child() obtains the devtree_lock and then calls
> of_device_is_available() which also attempts to claim the lock. This is
> obviously incorrect and causes a deadlock on boot. Fix issue by adding
> an variant of of_device_is_available() which doesn't obtain the lock.
> 
> Cc: Stephen Warren <swarren@wwwdotorg.org>
> Cc: Rob Herring <robherring2@gmail.com>
> Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
> ---
>  drivers/of/base.c |   15 +++++++++++++--
>  1 file changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/of/base.c b/drivers/of/base.c
> index e8d65af..4af74b7 100644
> --- a/drivers/of/base.c
> +++ b/drivers/of/base.c
> @@ -297,7 +297,7 @@ EXPORT_SYMBOL(of_machine_is_compatible);
>   *  Returns 1 if the status property is absent or set to "okay" or "ok",
>   *  0 otherwise
>   */
> -int of_device_is_available(const struct device_node *device)
> +int __of_device_is_available(const struct device_node *device)
>  {
>  	const char *status;
>  	int statlen;
> @@ -313,6 +313,17 @@ int of_device_is_available(const struct device_node *device)
>  
>  	return 0;
>  }
> +
> +int of_device_is_available(const struct device_node *device)

Stephen's version wins because he preserved the documentation. :)

Rob

> +{
> +	unsigned long flags;
> +	int rc;
> +
> +	raw_spin_lock_irqsave(&devtree_lock, flags);
> +	rc = __of_device_is_available(device);
> +	raw_spin_unlock_irqrestore(&devtree_lock, flags);
> +	return rc;
> +}
>  EXPORT_SYMBOL(of_device_is_available);
>  
>  /**
> @@ -404,7 +415,7 @@ struct device_node *of_get_next_available_child(const struct device_node *node,
>  	raw_spin_lock(&devtree_lock);
>  	next = prev ? prev->sibling : node->child;
>  	for (; next; next = next->sibling) {
> -		if (!of_device_is_available(next))
> +		if (!__of_device_is_available(next))
>  			continue;
>  		if (of_node_get(next))
>  			break;
> 

Re: [PATCH] of: Fix locking when calling of_get_next_available_child()

[Grant Likely]

On Mon, Feb 11, 2013 at 10:27 PM, Rob Herring <robherring2@gmail.com> wrote:
> On 02/11/2013 04:19 PM, Grant Likely wrote:
>> of_get_next_available_child() obtains the devtree_lock and then calls
>> of_device_is_available() which also attempts to claim the lock. This is
>> obviously incorrect and causes a deadlock on boot. Fix issue by adding
>> an variant of of_device_is_available() which doesn't obtain the lock.
>>
>> Cc: Stephen Warren <swarren@wwwdotorg.org>
>> Cc: Rob Herring <robherring2@gmail.com>
>> Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
>> ---
>>  drivers/of/base.c |   15 +++++++++++++--
>>  1 file changed, 13 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/of/base.c b/drivers/of/base.c
>> index e8d65af..4af74b7 100644
>> --- a/drivers/of/base.c
>> +++ b/drivers/of/base.c
>> @@ -297,7 +297,7 @@ EXPORT_SYMBOL(of_machine_is_compatible);
>>   *  Returns 1 if the status property is absent or set to "okay" or "ok",
>>   *  0 otherwise
>>   */
>> -int of_device_is_available(const struct device_node *device)
>> +int __of_device_is_available(const struct device_node *device)
>>  {
>>       const char *status;
>>       int statlen;
>> @@ -313,6 +313,17 @@ int of_device_is_available(const struct device_node *device)
>>
>>       return 0;
>>  }
>> +
>> +int of_device_is_available(const struct device_node *device)
>
> Stephen's version wins because he preserved the documentation. :)

Pish. He just cut and paste the documentation so now we have 2 copies.
My version wins because it is already in my tree. So ptfff!

:-)

g.

Re: [PATCH] of: Fix locking when calling of_get_next_available_child()

[Stephen Warren]

On 02/11/2013 03:19 PM, Grant Likely wrote:
> of_get_next_available_child() obtains the devtree_lock and then calls
> of_device_is_available() which also attempts to claim the lock. This is
> obviously incorrect and causes a deadlock on boot. Fix issue by adding
> an variant of of_device_is_available() which doesn't obtain the lock.

This patch forgets to update __of_device_is_available() to call
__of_get_property() rather than of_get_property() and hence doesn't
actually solve the problem.

You might want to make __of_device_is_available() static too.

Re: [PATCH] of: Fix locking when calling of_get_next_available_child()

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 739 bytes --]

On Mon, Feb 11, 2013 at 04:26:46PM -0700, Stephen Warren wrote:
> On 02/11/2013 03:19 PM, Grant Likely wrote:
> > of_get_next_available_child() obtains the devtree_lock and then calls
> > of_device_is_available() which also attempts to claim the lock. This is
> > obviously incorrect and causes a deadlock on boot. Fix issue by adding
> > an variant of of_device_is_available() which doesn't obtain the lock.
> 
> This patch forgets to update __of_device_is_available() to call
> __of_get_property() rather than of_get_property() and hence doesn't
> actually solve the problem.

Yes, please. of_get_property() called from __of_device_is_available()
pretty much defeats the purpose of having the non-locked version.

Thierry

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH] of: Fix locking when calling of_get_next_available_child()

[Grant Likely]

On Mon, 11 Feb 2013 16:26:46 -0700, Stephen Warren <swarren@wwwdotorg.org> wrote:
> On 02/11/2013 03:19 PM, Grant Likely wrote:
> > of_get_next_available_child() obtains the devtree_lock and then calls
> > of_device_is_available() which also attempts to claim the lock. This is
> > obviously incorrect and causes a deadlock on boot. Fix issue by adding
> > an variant of of_device_is_available() which doesn't obtain the lock.
> 
> This patch forgets to update __of_device_is_available() to call
> __of_get_property() rather than of_get_property() and hence doesn't
> actually solve the problem.
> 
> You might want to make __of_device_is_available() static too.

Okay, I've dropped this patch and replaced it with your version since
you're actually doing better testing that I it would appear. I had to
respin my tree anyway because I messed up a merge of Rob's branch. :-(
Sorry for the noise.

New version of the tree is pushed out.

g.