[PATCH v2] usb: roles: add lockdep class key to struct usb_role_switch

[PATCH v2] usb: roles: add lockdep class key to struct usb_role_switch

[Amit Sunil Dhamne]

There can be multiple role switch devices running on a platform. Given
that lockdep is not capable of differentiating between locks of
different instances, false positive warnings for circular locking are
reported. To prevent this, register unique lockdep key for each of the
individual instances.

Signed-off-by: Amit Sunil Dhamne <amitsd@google.com>
---
v1->v2
- Avoid usage of ifdefs.
---
 drivers/usb/roles/class.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/usb/roles/class.c b/drivers/usb/roles/class.c
index d7aa913ceb8a..9dbe9f6bea83 100644
--- a/drivers/usb/roles/class.c
+++ b/drivers/usb/roles/class.c
@@ -11,6 +11,7 @@
 #include <linux/usb/role.h>
 #include <linux/property.h>
 #include <linux/device.h>
+#include <linux/lockdep.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/slab.h>
@@ -33,6 +34,8 @@ struct usb_role_switch {
 	usb_role_switch_set_t set;
 	usb_role_switch_get_t get;
 	bool allow_userspace_control;
+
+	struct lock_class_key key;
 };
 
 #define to_role_switch(d)	container_of(d, struct usb_role_switch, dev)
@@ -396,6 +399,11 @@ usb_role_switch_register(struct device *parent,
 
 	sw->registered = true;
 
+	if (IS_ENABLED(CONFIG_LOCKDEP)) {
+		lockdep_register_key(&sw->key);
+		lockdep_set_class(&sw->lock, &sw->key);
+	}
+
 	/* TODO: Symlinks for the host port and the device controller. */
 
 	return sw;
@@ -412,6 +420,10 @@ void usb_role_switch_unregister(struct usb_role_switch *sw)
 {
 	if (IS_ERR_OR_NULL(sw))
 		return;
+
+	if (IS_ENABLED(CONFIG_LOCKDEP))
+		lockdep_unregister_key(&sw->key);
+
 	sw->registered = false;
 	if (dev_fwnode(&sw->dev))
 		component_del(&sw->dev, &connector_ops);
-- 
2.46.0.76.ge559c4bf1a-goog

Re: [PATCH v2] usb: roles: add lockdep class key to struct usb_role_switch

[Greg KH]

On Wed, Aug 14, 2024 at 09:40:55PM -0700, Amit Sunil Dhamne wrote:
> There can be multiple role switch devices running on a platform. Given
> that lockdep is not capable of differentiating between locks of
> different instances, false positive warnings for circular locking are
> reported. To prevent this, register unique lockdep key for each of the
> individual instances.
> 
> Signed-off-by: Amit Sunil Dhamne <amitsd@google.com>
> ---
> v1->v2
> - Avoid usage of ifdefs.
> ---
>  drivers/usb/roles/class.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/drivers/usb/roles/class.c b/drivers/usb/roles/class.c
> index d7aa913ceb8a..9dbe9f6bea83 100644
> --- a/drivers/usb/roles/class.c
> +++ b/drivers/usb/roles/class.c
> @@ -11,6 +11,7 @@
>  #include <linux/usb/role.h>
>  #include <linux/property.h>
>  #include <linux/device.h>
> +#include <linux/lockdep.h>
>  #include <linux/module.h>
>  #include <linux/mutex.h>
>  #include <linux/slab.h>
> @@ -33,6 +34,8 @@ struct usb_role_switch {
>  	usb_role_switch_set_t set;
>  	usb_role_switch_get_t get;
>  	bool allow_userspace_control;
> +
> +	struct lock_class_key key;
>  };
>  
>  #define to_role_switch(d)	container_of(d, struct usb_role_switch, dev)
> @@ -396,6 +399,11 @@ usb_role_switch_register(struct device *parent,
>  
>  	sw->registered = true;
>  
> +	if (IS_ENABLED(CONFIG_LOCKDEP)) {

Why is this if statement needed at all?

Please get someone internal to your company/group to review the change
before sending it out for others to find the obvious issues with it.
That's what your peers are for.  Please do so and get them to provide a
reviewed-by on it before submitting it again.

thanks,

greg k-h

Re: [PATCH v2] usb: roles: add lockdep class key to struct usb_role_switch

[Amit Sunil Dhamne]

On 8/14/24 10:06 PM, Greg KH wrote:
> On Wed, Aug 14, 2024 at 09:40:55PM -0700, Amit Sunil Dhamne wrote:
>> There can be multiple role switch devices running on a platform. Given
>> that lockdep is not capable of differentiating between locks of
>> different instances, false positive warnings for circular locking are
>> reported. To prevent this, register unique lockdep key for each of the
>> individual instances.
>>
>> Signed-off-by: Amit Sunil Dhamne <amitsd@google.com>
>> ---
>> v1->v2
>> - Avoid usage of ifdefs.
>> ---
>>   drivers/usb/roles/class.c | 12 ++++++++++++
>>   1 file changed, 12 insertions(+)
>>
>> diff --git a/drivers/usb/roles/class.c b/drivers/usb/roles/class.c
>> index d7aa913ceb8a..9dbe9f6bea83 100644
>> --- a/drivers/usb/roles/class.c
>> +++ b/drivers/usb/roles/class.c
>> @@ -11,6 +11,7 @@
>>   #include <linux/usb/role.h>
>>   #include <linux/property.h>
>>   #include <linux/device.h>
>> +#include <linux/lockdep.h>
>>   #include <linux/module.h>
>>   #include <linux/mutex.h>
>>   #include <linux/slab.h>
>> @@ -33,6 +34,8 @@ struct usb_role_switch {
>>   	usb_role_switch_set_t set;
>>   	usb_role_switch_get_t get;
>>   	bool allow_userspace_control;
>> +
>> +	struct lock_class_key key;
>>   };
>>   
>>   #define to_role_switch(d)	container_of(d, struct usb_role_switch, dev)
>> @@ -396,6 +399,11 @@ usb_role_switch_register(struct device *parent,
>>   
>>   	sw->registered = true;
>>   
>> +	if (IS_ENABLED(CONFIG_LOCKDEP)) {
> Why is this if statement needed at all?

Upon closer look it's redundant. Unfortunately, I missed the different 
definitions for the lockdep API & structs based on CONFIG_LOCKDEP in hdr 
file.

Will be more careful next time.

>
> Please get someone internal to your company/group to review the change
> before sending it out for others to find the obvious issues with it.
> That's what your peers are for.  Please do so and get them to provide a
> reviewed-by on it before submitting it again.

Sounds good. I will send the next revision with reviewed-by signature 
from a peer.


Thanks,

Amit

> thanks,
>
> greg k-h