[PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[Aaron Lu]

Commit 0090def("ACPI: Add interface to register/unregister device
to/from power resources") used resource_lock to protect the devices list
that relies on power resource. It caused a mutex dead lock, as below

    acpi_power_on ---> lock resource_lock
      __acpi_power_on
        acpi_power_on_device
          acpi_power_get_inferred_state
            acpi_power_get_list_state ---> lock resource_lock

This patch adds a new mutex "devices_lock" to protect the devices list
and calls acpi_power_on_device in acpi_power_on, instead of
__acpi_power_on, after the resource_lock is released.

Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
---
v2:
If power resource is already on, no need to check if device needs
to be resumed.
v1:
By Lin Ming.

 drivers/acpi/power.c | 36 ++++++++++++++++++++++++------------
 1 file changed, 24 insertions(+), 12 deletions(-)

diff --git a/drivers/acpi/power.c b/drivers/acpi/power.c
index 215ecd0..3582a26 100644
--- a/drivers/acpi/power.c
+++ b/drivers/acpi/power.c
@@ -105,6 +105,7 @@ struct acpi_power_resource {
 
 	/* List of devices relying on this power resource */
 	struct acpi_power_resource_device *devices;
+	struct mutex devices_lock;
 };
 
 static struct list_head acpi_power_resource_list;
@@ -223,7 +224,6 @@ static void acpi_power_on_device(struct acpi_power_managed_device *device)
 
 static int __acpi_power_on(struct acpi_power_resource *resource)
 {
-	struct acpi_power_resource_device *device_list = resource->devices;
 	acpi_status status = AE_OK;
 
 	status = acpi_evaluate_object(resource->device->handle, "_ON", NULL, NULL);
@@ -236,19 +236,14 @@ static int __acpi_power_on(struct acpi_power_resource *resource)
 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Power resource [%s] turned on\n",
 			  resource->name));
 
-	while (device_list) {
-		acpi_power_on_device(device_list->device);
-
-		device_list = device_list->next;
-	}
-
 	return 0;
 }
 
 static int acpi_power_on(acpi_handle handle)
 {
-	int result = 0;
+	int result = 0, resume_device = 0;
 	struct acpi_power_resource *resource = NULL;
+	struct acpi_power_resource_device *device_list;
 
 	result = acpi_power_get_context(handle, &resource);
 	if (result)
@@ -264,10 +259,26 @@ static int acpi_power_on(acpi_handle handle)
 		result = __acpi_power_on(resource);
 		if (result)
 			resource->ref_count--;
+		else
+			resume_device = 1;
 	}
 
 	mutex_unlock(&resource->resource_lock);
 
+	if (!resume_device)
+		return result;
+
+	mutex_lock(&resource->devices_lock);
+
+	device_list = resource->devices;
+	while (device_list) {
+		acpi_power_on_device(device_list->device);
+
+		device_list = device_list->next;
+	}
+
+	mutex_unlock(&resource->devices_lock);
+
 	return result;
 }
 
@@ -353,7 +364,7 @@ static void __acpi_power_resource_unregister_device(struct device *dev,
 	if (acpi_power_get_context(res_handle, &resource))
 		return;
 
-	mutex_lock(&resource->resource_lock);
+	mutex_lock(&resource->devices_lock);
 	prev = NULL;
 	curr = resource->devices;
 	while (curr) {
@@ -370,7 +381,7 @@ static void __acpi_power_resource_unregister_device(struct device *dev,
 		prev = curr;
 		curr = curr->next;
 	}
-	mutex_unlock(&resource->resource_lock);
+	mutex_unlock(&resource->devices_lock);
 }
 
 /* Unlink dev from all power resources in _PR0 */
@@ -412,10 +423,10 @@ static int __acpi_power_resource_register_device(
 
 	power_resource_device->device = powered_device;
 
-	mutex_lock(&resource->resource_lock);
+	mutex_lock(&resource->devices_lock);
 	power_resource_device->next = resource->devices;
 	resource->devices = power_resource_device;
-	mutex_unlock(&resource->resource_lock);
+	mutex_unlock(&resource->devices_lock);
 
 	return 0;
 }
@@ -719,6 +730,7 @@ static int acpi_power_add(struct acpi_device *device)
 
 	resource->device = device;
 	mutex_init(&resource->resource_lock);
+	mutex_init(&resource->devices_lock);
 	strcpy(resource->name, device->pnp.bus_id);
 	strcpy(acpi_device_name(device), ACPI_POWER_DEVICE_NAME);
 	strcpy(acpi_device_class(device), ACPI_POWER_CLASS);
-- 
1.7.12.21.g871e293

Re: [PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[Rafael J. Wysocki]

On Wednesday, September 12, 2012, Aaron Lu wrote:
> Commit 0090def("ACPI: Add interface to register/unregister device
> to/from power resources") used resource_lock to protect the devices list
> that relies on power resource. It caused a mutex dead lock, as below
> 
>     acpi_power_on ---> lock resource_lock
>       __acpi_power_on
>         acpi_power_on_device
>           acpi_power_get_inferred_state
>             acpi_power_get_list_state ---> lock resource_lock
> 
> This patch adds a new mutex "devices_lock" to protect the devices list
> and calls acpi_power_on_device in acpi_power_on, instead of
> __acpi_power_on, after the resource_lock is released.
> 
> Signed-off-by: Lin Ming <ming.m.lin@intel.com>
> Signed-off-by: Aaron Lu <aaron.lu@intel.com>

Good catch, thanks.

I hope Len won't mind if I take it for v3.6.

> ---
> v2:
> If power resource is already on, no need to check if device needs
> to be resumed.
> v1:
> By Lin Ming.
> 
>  drivers/acpi/power.c | 36 ++++++++++++++++++++++++------------
>  1 file changed, 24 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/acpi/power.c b/drivers/acpi/power.c
> index 215ecd0..3582a26 100644
> --- a/drivers/acpi/power.c
> +++ b/drivers/acpi/power.c
> @@ -105,6 +105,7 @@ struct acpi_power_resource {
>  
>  	/* List of devices relying on this power resource */
>  	struct acpi_power_resource_device *devices;
> +	struct mutex devices_lock;
>  };
>  
>  static struct list_head acpi_power_resource_list;
> @@ -223,7 +224,6 @@ static void acpi_power_on_device(struct acpi_power_managed_device *device)
>  
>  static int __acpi_power_on(struct acpi_power_resource *resource)
>  {
> -	struct acpi_power_resource_device *device_list = resource->devices;
>  	acpi_status status = AE_OK;
>  
>  	status = acpi_evaluate_object(resource->device->handle, "_ON", NULL, NULL);
> @@ -236,19 +236,14 @@ static int __acpi_power_on(struct acpi_power_resource *resource)
>  	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Power resource [%s] turned on\n",
>  			  resource->name));
>  
> -	while (device_list) {
> -		acpi_power_on_device(device_list->device);
> -
> -		device_list = device_list->next;
> -	}
> -
>  	return 0;
>  }
>  
>  static int acpi_power_on(acpi_handle handle)
>  {
> -	int result = 0;
> +	int result = 0, resume_device = 0;

I'll change the data type of resume_device to bool when applying the patch.

>  	struct acpi_power_resource *resource = NULL;
> +	struct acpi_power_resource_device *device_list;
>  
>  	result = acpi_power_get_context(handle, &resource);
>  	if (result)
> @@ -264,10 +259,26 @@ static int acpi_power_on(acpi_handle handle)
>  		result = __acpi_power_on(resource);
>  		if (result)
>  			resource->ref_count--;
> +		else
> +			resume_device = 1;
>  	}
>  
>  	mutex_unlock(&resource->resource_lock);
>  
> +	if (!resume_device)
> +		return result;
> +
> +	mutex_lock(&resource->devices_lock);
> +
> +	device_list = resource->devices;
> +	while (device_list) {
> +		acpi_power_on_device(device_list->device);
> +
> +		device_list = device_list->next;
> +	}
> +
> +	mutex_unlock(&resource->devices_lock);
> +
>  	return result;
>  }
>  
> @@ -353,7 +364,7 @@ static void __acpi_power_resource_unregister_device(struct device *dev,
>  	if (acpi_power_get_context(res_handle, &resource))
>  		return;
>  
> -	mutex_lock(&resource->resource_lock);
> +	mutex_lock(&resource->devices_lock);
>  	prev = NULL;
>  	curr = resource->devices;
>  	while (curr) {
> @@ -370,7 +381,7 @@ static void __acpi_power_resource_unregister_device(struct device *dev,
>  		prev = curr;
>  		curr = curr->next;
>  	}
> -	mutex_unlock(&resource->resource_lock);
> +	mutex_unlock(&resource->devices_lock);
>  }
>  
>  /* Unlink dev from all power resources in _PR0 */
> @@ -412,10 +423,10 @@ static int __acpi_power_resource_register_device(
>  
>  	power_resource_device->device = powered_device;
>  
> -	mutex_lock(&resource->resource_lock);
> +	mutex_lock(&resource->devices_lock);
>  	power_resource_device->next = resource->devices;
>  	resource->devices = power_resource_device;
> -	mutex_unlock(&resource->resource_lock);
> +	mutex_unlock(&resource->devices_lock);
>  
>  	return 0;
>  }
> @@ -719,6 +730,7 @@ static int acpi_power_add(struct acpi_device *device)
>  
>  	resource->device = device;
>  	mutex_init(&resource->resource_lock);
> +	mutex_init(&resource->devices_lock);
>  	strcpy(resource->name, device->pnp.bus_id);
>  	strcpy(acpi_device_name(device), ACPI_POWER_DEVICE_NAME);
>  	strcpy(acpi_device_class(device), ACPI_POWER_CLASS);
> 

Thanks,
Rafael

Re: [PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[Aaron Lu]

On 09/14/2012 06:26 AM, Rafael J. Wysocki wrote:
> On Wednesday, September 12, 2012, Aaron Lu wrote:
>> Commit 0090def("ACPI: Add interface to register/unregister device
>> to/from power resources") used resource_lock to protect the devices list
>> that relies on power resource. It caused a mutex dead lock, as below
>>
>>     acpi_power_on ---> lock resource_lock
>>       __acpi_power_on
>>         acpi_power_on_device
>>           acpi_power_get_inferred_state
>>             acpi_power_get_list_state ---> lock resource_lock
>>
>> This patch adds a new mutex "devices_lock" to protect the devices list
>> and calls acpi_power_on_device in acpi_power_on, instead of
>> __acpi_power_on, after the resource_lock is released.
>>
>> Signed-off-by: Lin Ming <ming.m.lin@intel.com>
>> Signed-off-by: Aaron Lu <aaron.lu@intel.com>
> 
> Good catch, thanks.
> 
> I hope Len won't mind if I take it for v3.6.
> 

Yes, that would be good.
And the commit enters v3.4, so a stable tag may need be added.

>> ---
>> v2:
>> If power resource is already on, no need to check if device needs
>> to be resumed.
>> v1:
>> By Lin Ming.
>>
>>  drivers/acpi/power.c | 36 ++++++++++++++++++++++++------------
>>  1 file changed, 24 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/acpi/power.c b/drivers/acpi/power.c
>> index 215ecd0..3582a26 100644
>> --- a/drivers/acpi/power.c
>> +++ b/drivers/acpi/power.c
>> @@ -105,6 +105,7 @@ struct acpi_power_resource {
>>  
>>  	/* List of devices relying on this power resource */
>>  	struct acpi_power_resource_device *devices;
>> +	struct mutex devices_lock;
>>  };
>>  
>>  static struct list_head acpi_power_resource_list;
>> @@ -223,7 +224,6 @@ static void acpi_power_on_device(struct acpi_power_managed_device *device)
>>  
>>  static int __acpi_power_on(struct acpi_power_resource *resource)
>>  {
>> -	struct acpi_power_resource_device *device_list = resource->devices;
>>  	acpi_status status = AE_OK;
>>  
>>  	status = acpi_evaluate_object(resource->device->handle, "_ON", NULL, NULL);
>> @@ -236,19 +236,14 @@ static int __acpi_power_on(struct acpi_power_resource *resource)
>>  	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Power resource [%s] turned on\n",
>>  			  resource->name));
>>  
>> -	while (device_list) {
>> -		acpi_power_on_device(device_list->device);
>> -
>> -		device_list = device_list->next;
>> -	}
>> -
>>  	return 0;
>>  }
>>  
>>  static int acpi_power_on(acpi_handle handle)
>>  {
>> -	int result = 0;
>> +	int result = 0, resume_device = 0;
> 
> I'll change the data type of resume_device to bool when applying the patch.
> 

Sure, thanks.

-Aaron

Re: [PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[Aaron Lu]

On 09/14/2012 09:15 AM, Aaron Lu wrote:
>>
>> Good catch, thanks.
>>
>> I hope Len won't mind if I take it for v3.6.
>>
> 
> Yes, that would be good.
> And the commit enters v3.4, so a stable tag may need be added.
> 

On another thought, it may not be needed for pre v3.6 kernel as the only
code that uses this interface appeared in v3.6-rc1.

Thanks,
Aaron

Re: [PATCH v2] ACPI: Fix resource_lock dead lock in acpi_power_on_device

[Rafael J. Wysocki]

On Friday, September 14, 2012, Aaron Lu wrote:
> On 09/14/2012 09:15 AM, Aaron Lu wrote:
> >>
> >> Good catch, thanks.
> >>
> >> I hope Len won't mind if I take it for v3.6.
> >>
> > 
> > Yes, that would be good.
> > And the commit enters v3.4, so a stable tag may need be added.
> > 
> 
> On another thought, it may not be needed for pre v3.6 kernel as the only
> code that uses this interface appeared in v3.6-rc1.

I've added the -stable tag already.  It may be useful for backports and the
bug is real.

Thanks,
Rafael