[PATCH] thermal: bcm2835: Fix crash in bcm2835_thermal_debugfs

[PATCH] thermal: bcm2835: Fix crash in bcm2835_thermal_debugfs

[Phil Elwell]

"cat /sys/kernel/debug/bcm2835_thermal/regset" causes a NULL pointer
dereference in bcm2835_thermal_debugfs. The driver makes use of the
implementation details of the thermal framework to retrieve a pointer
to its private data from a struct thermal_zone_device, and gets it
wrong - leading to the crash. Instead, store its private data as the
drvdata and retrieve the thermal_zone_device pointer from it.

Fixes: bcb7dd9ef206 ("thermal: bcm2835: add thermal driver for bcm2835 SoC")

Signed-off-by: Phil Elwell <phil@raspberrypi.org>
---
 drivers/thermal/broadcom/bcm2835_thermal.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/drivers/thermal/broadcom/bcm2835_thermal.c b/drivers/thermal/broadcom/bcm2835_thermal.c
index 720760c..ba39647 100644
--- a/drivers/thermal/broadcom/bcm2835_thermal.c
+++ b/drivers/thermal/broadcom/bcm2835_thermal.c
@@ -119,8 +119,7 @@ static const struct debugfs_reg32 bcm2835_thermal_regs[] = {
 
 static void bcm2835_thermal_debugfs(struct platform_device *pdev)
 {
-	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
-	struct bcm2835_thermal_data *data = tz->devdata;
+	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
 	struct debugfs_regset32 *regset;
 
 	data->debugfsdir = debugfs_create_dir("bcm2835_thermal", NULL);
@@ -266,7 +265,7 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
 
 	data->tz = tz;
 
-	platform_set_drvdata(pdev, tz);
+	platform_set_drvdata(pdev, data);
 
 	/*
 	 * Thermal_zone doesn't enable hwmon as default,
@@ -290,8 +289,8 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
 
 static int bcm2835_thermal_remove(struct platform_device *pdev)
 {
-	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
-	struct bcm2835_thermal_data *data = tz->devdata;
+	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
+	struct thermal_zone_device *tz = data->tz;
 
 	debugfs_remove_recursive(data->debugfsdir);
 	thermal_zone_of_sensor_unregister(&pdev->dev, tz);
-- 
2.7.4

Re: [PATCH] thermal: bcm2835: Fix crash in bcm2835_thermal_debugfs

[Daniel Lezcano]

On 29/01/2019 10:10, Phil Elwell wrote:
> "cat /sys/kernel/debug/bcm2835_thermal/regset" causes a NULL pointer
> dereference in bcm2835_thermal_debugfs. The driver makes use of the
> implementation details of the thermal framework to retrieve a pointer
> to its private data from a struct thermal_zone_device, and gets it
> wrong - leading to the crash. Instead, store its private data as the
> drvdata and retrieve the thermal_zone_device pointer from it.
> 
> Fixes: bcb7dd9ef206 ("thermal: bcm2835: add thermal driver for bcm2835 SoC")
> 
> Signed-off-by: Phil Elwell <phil@raspberrypi.org>

Reviewed-by: Daniel Lezcano <daniel.lezcano@linaro.org>

One question: do you really want the thermal driver to be loaded as a
module ? Don't you want to have it operational as soon as possible
instead of waiting the OS to boot and load the kernel module?

If a module format is not needed, then the thermal_zone field won't be
needed anymore because the 'remove' ops can be removed and the need of
the tz field to unregister it disappears.

> ---
>  drivers/thermal/broadcom/bcm2835_thermal.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/thermal/broadcom/bcm2835_thermal.c b/drivers/thermal/broadcom/bcm2835_thermal.c
> index 720760c..ba39647 100644
> --- a/drivers/thermal/broadcom/bcm2835_thermal.c
> +++ b/drivers/thermal/broadcom/bcm2835_thermal.c
> @@ -119,8 +119,7 @@ static const struct debugfs_reg32 bcm2835_thermal_regs[] = {
>  
>  static void bcm2835_thermal_debugfs(struct platform_device *pdev)
>  {
> -	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
> -	struct bcm2835_thermal_data *data = tz->devdata;
> +	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
>  	struct debugfs_regset32 *regset;
>  
>  	data->debugfsdir = debugfs_create_dir("bcm2835_thermal", NULL);
> @@ -266,7 +265,7 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
>  
>  	data->tz = tz;
>  
> -	platform_set_drvdata(pdev, tz);
> +	platform_set_drvdata(pdev, data);
>  
>  	/*
>  	 * Thermal_zone doesn't enable hwmon as default,
> @@ -290,8 +289,8 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
>  
>  static int bcm2835_thermal_remove(struct platform_device *pdev)
>  {
> -	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
> -	struct bcm2835_thermal_data *data = tz->devdata;
> +	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
> +	struct thermal_zone_device *tz = data->tz;
>  
>  	debugfs_remove_recursive(data->debugfsdir);
>  	thermal_zone_of_sensor_unregister(&pdev->dev, tz);
> 


-- 
 <http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

Re: [PATCH] thermal: bcm2835: Fix crash in bcm2835_thermal_debugfs

[Phil Elwell]

Hi Daniel,

On 29/01/2019 09:52, Daniel Lezcano wrote:
> On 29/01/2019 10:10, Phil Elwell wrote:
>> "cat /sys/kernel/debug/bcm2835_thermal/regset" causes a NULL pointer
>> dereference in bcm2835_thermal_debugfs. The driver makes use of the
>> implementation details of the thermal framework to retrieve a pointer
>> to its private data from a struct thermal_zone_device, and gets it
>> wrong - leading to the crash. Instead, store its private data as the
>> drvdata and retrieve the thermal_zone_device pointer from it.
>>
>> Fixes: bcb7dd9ef206 ("thermal: bcm2835: add thermal driver for bcm2835 SoC")
>>
>> Signed-off-by: Phil Elwell <phil@raspberrypi.org>
> 
> Reviewed-by: Daniel Lezcano <daniel.lezcano@linaro.org>
> 
> One question: do you really want the thermal driver to be loaded as a
> module ? Don't you want to have it operational as soon as possible
> instead of waiting the OS to boot and load the kernel module?
> 
> If a module format is not needed, then the thermal_zone field won't be
> needed anymore because the 'remove' ops can be removed and the need of
> the tz field to unregister it disappears.

Thanks for the review. The general approach for Pi drivers has been to make them
tristate to allow minimal kernels to be built - important when the smallest
Pis have 256M - and the delayed startup isn't a problem because the firmware
is running a thermal limiter from before the kernel is loaded. However, this
driver is very small, and I wouldn't object if somebody decided it was better
to make it built-in.

> 
>> ---
>>  drivers/thermal/broadcom/bcm2835_thermal.c | 9 ++++-----
>>  1 file changed, 4 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/thermal/broadcom/bcm2835_thermal.c b/drivers/thermal/broadcom/bcm2835_thermal.c
>> index 720760c..ba39647 100644
>> --- a/drivers/thermal/broadcom/bcm2835_thermal.c
>> +++ b/drivers/thermal/broadcom/bcm2835_thermal.c
>> @@ -119,8 +119,7 @@ static const struct debugfs_reg32 bcm2835_thermal_regs[] = {
>>  
>>  static void bcm2835_thermal_debugfs(struct platform_device *pdev)
>>  {
>> -	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
>> -	struct bcm2835_thermal_data *data = tz->devdata;
>> +	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
>>  	struct debugfs_regset32 *regset;
>>  
>>  	data->debugfsdir = debugfs_create_dir("bcm2835_thermal", NULL);
>> @@ -266,7 +265,7 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
>>  
>>  	data->tz = tz;
>>  
>> -	platform_set_drvdata(pdev, tz);
>> +	platform_set_drvdata(pdev, data);
>>  
>>  	/*
>>  	 * Thermal_zone doesn't enable hwmon as default,
>> @@ -290,8 +289,8 @@ static int bcm2835_thermal_probe(struct platform_device *pdev)
>>  
>>  static int bcm2835_thermal_remove(struct platform_device *pdev)
>>  {
>> -	struct thermal_zone_device *tz = platform_get_drvdata(pdev);
>> -	struct bcm2835_thermal_data *data = tz->devdata;
>> +	struct bcm2835_thermal_data *data = platform_get_drvdata(pdev);
>> +	struct thermal_zone_device *tz = data->tz;
>>  
>>  	debugfs_remove_recursive(data->debugfsdir);
>>  	thermal_zone_of_sensor_unregister(&pdev->dev, tz);
>>
> 
> 

Re: [PATCH] thermal: bcm2835: Fix crash in bcm2835_thermal_debugfs

[Phil Elwell]

Hi Stefan,

On 29/01/2019 09:44, Stefan Wahren wrote:
> Hi Phil,
> 
> Am 29.01.2019 um 10:10 schrieb Phil Elwell:
>> "cat /sys/kernel/debug/bcm2835_thermal/regset" causes a NULL pointer
>> dereference in bcm2835_thermal_debugfs. The driver makes use of the
>> implementation details of the thermal framework to retrieve a pointer
>> to its private data from a struct thermal_zone_device, and gets it
>> wrong - leading to the crash. Instead, store its private data as the
>> drvdata and retrieve the thermal_zone_device pointer from it.
>>
>> Fixes: bcb7dd9ef206 ("thermal: bcm2835: add thermal driver for bcm2835 SoC")
>>
>> Signed-off-by: Phil Elwell <phil-FnsA7b+Nu9XbIbC87yuRow@public.gmane.org>
> 
> thanks for this patch.
> 
> Could you please resend this patch to the thermal maintainers Zhang Rui and Eduardo Valentin?||||
> 
> Regards
> Stefan
> 

I blame the get_maintainer script - a "Supporter" sounds less invested and less authoritative
than a "Maintainer".

A resend is imminent.

Phil