Re: [PATCH v3 4/4] drivers/thermal/exymos: Use guard notation when acquiring mutex

[Lukasz Luba <lukasz.luba@arm.com>]

On 3/5/25 15:59, Anand Moon wrote:
> Hi Lukasz,
> 
> On Wed, 5 Mar 2025 at 14:12, Lukasz Luba <lukasz.luba@arm.com> wrote:
>>
>>
>>
>> On 3/4/25 12:20, Anand Moon wrote:
>>> Hi Lukasz,
>>>
>>> On Sat, 1 Mar 2025 at 00:06, Anand Moon <linux.amoon@gmail.com> wrote:
>>>>
>>>> Hi Lukasz,
>>>>
>>>> On Fri, 28 Feb 2025 at 22:58, Lukasz Luba <lukasz.luba@arm.com> wrote:
>>>>>
>>>>>
>>>>>
>>>>> On 2/16/25 19:58, Anand Moon wrote:
>>>>>> Using guard notation makes the code more compact and error handling
>>>>>> more robust by ensuring that mutexes are released in all code paths
>>>>>> when control leaves critical section.
>>>>>>
>>>>>> Signed-off-by: Anand Moon <linux.amoon@gmail.com>
>>>>>> ---
>>>>>> v3: new patch
>>>>>> ---
>>>>>>     drivers/thermal/samsung/exynos_tmu.c | 21 +++++++--------------
>>>>>>     1 file changed, 7 insertions(+), 14 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/thermal/samsung/exynos_tmu.c b/drivers/thermal/samsung/exynos_tmu.c
>>>>>> index fe090c1a93ab..a34ba3858d64 100644
>>>>>> --- a/drivers/thermal/samsung/exynos_tmu.c
>>>>>> +++ b/drivers/thermal/samsung/exynos_tmu.c
>>>>>> @@ -256,7 +256,7 @@ static int exynos_tmu_initialize(struct platform_device *pdev)
>>>>>>         unsigned int status;
>>>>>>         int ret = 0;
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>         clk_enable(data->clk_sec);
>>>>>>
>>>>>> @@ -270,7 +270,6 @@ static int exynos_tmu_initialize(struct platform_device *pdev)
>>>>>>
>>>>>>         clk_disable(data->clk_sec);
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>
>>>>>>         return ret;
>>>>>>     }
>>>>>> @@ -292,13 +291,12 @@ static int exynos_thermal_zone_configure(struct platform_device *pdev)
>>>>>>                 return ret;
>>>>>>         }
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>
>>>>>>         data->tmu_set_crit_temp(data, temp / MCELSIUS);
>>>>>>
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>
>>>>>>         return 0;
>>>>>>     }
>>>>>> @@ -325,12 +323,11 @@ static void exynos_tmu_control(struct platform_device *pdev, bool on)
>>>>>>     {
>>>>>>         struct exynos_tmu_data *data = platform_get_drvdata(pdev);
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>         data->tmu_control(pdev, on);
>>>>>>         data->enabled = on;
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>     }
>>>>>>
>>>>>>     static void exynos_tmu_update_bit(struct exynos_tmu_data *data, int reg_off,
>>>>>> @@ -645,7 +642,7 @@ static int exynos_get_temp(struct thermal_zone_device *tz, int *temp)
>>>>>>                  */
>>>>>>                 return -EAGAIN;
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>
>>>>>>         value = data->tmu_read(data);
>>>>>> @@ -655,7 +652,6 @@ static int exynos_get_temp(struct thermal_zone_device *tz, int *temp)
>>>>>>                 *temp = code_to_temp(data, value) * MCELSIUS;
>>>>>>
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>
>>>>>>         return ret;
>>>>>>     }
>>>>>> @@ -720,11 +716,10 @@ static int exynos_tmu_set_emulation(struct thermal_zone_device *tz, int temp)
>>>>>>         if (temp && temp < MCELSIUS)
>>>>>>                 goto out;
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>         data->tmu_set_emulation(data, temp);
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>         return 0;
>>>>>>     out:
>>>>>>         return ret;
>>>>>> @@ -760,14 +755,13 @@ static irqreturn_t exynos_tmu_threaded_irq(int irq, void *id)
>>>>>>
>>>>>>         thermal_zone_device_update(data->tzd, THERMAL_EVENT_UNSPECIFIED);
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>
>>>>>>         /* TODO: take action based on particular interrupt */
>>>>>>         data->tmu_clear_irqs(data);
>>>>>>
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>
>>>>>>         return IRQ_HANDLED;
>>>>>>     }
>>>>>> @@ -987,7 +981,7 @@ static int exynos_set_trips(struct thermal_zone_device *tz, int low, int high)
>>>>>>     {
>>>>>>         struct exynos_tmu_data *data = thermal_zone_device_priv(tz);
>>>>>>
>>>>>> -     mutex_lock(&data->lock);
>>>>>> +     guard(mutex)(&data->lock);
>>>>>>         clk_enable(data->clk);
>>>>>>
>>>>>>         if (low > INT_MIN)
>>>>>> @@ -1000,7 +994,6 @@ static int exynos_set_trips(struct thermal_zone_device *tz, int low, int high)
>>>>>>                 data->tmu_disable_high(data);
>>>>>>
>>>>>>         clk_disable(data->clk);
>>>>>> -     mutex_unlock(&data->lock);
>>>>>>
>>>>>>         return 0;
>>>>>>     }
>>>>
>>>> Thanks for your review comments.
>>>>>
>>>>> IMO you should be able to even use something like we have
>>>>> core framework:
>>>>>
>>>>> guard(thermal_zone)(tz);
>>>>>
>>>>> Your mutex name is simply 'lock' in the struct exynos_tmu_data
>>>>> so you should be able to leverage this by:
>>>>>
>>>>> guard(exynos_tmu_data)(data);
>>>>>
>>>
>>> If I introduce the guard it creates a compilation error
>>>
>>> amoon@anand-m920q:~/mainline/linux-exynos-6.y-devel$ vi
>>> drivers/thermal/samsung/exynos_tmu.c +306
>>> amoon@anand-m920q:~/mainline/linux-exynos-6.y-devel$ make -j$(nproc)
>>> ARCH=arm CROSS_COMPILE=arm-none-eabi- LOCALVERSION=-u3ml dtbs zImage
>>> modules
>>>     CALL    scripts/checksyscalls.sh
>>>     CHK     kernel/kheaders_data.tar.xz
>>>     CC      drivers/thermal/samsung/exynos_tmu.o
>>>     CC [M]  drivers/md/raid10.o
>>> In file included from ./include/linux/irqflags.h:17,
>>>                    from ./arch/arm/include/asm/bitops.h:28,
>>>                    from ./include/linux/bitops.h:68,
>>>                    from ./include/linux/kernel.h:23,
>>>                    from ./include/linux/clk.h:13,
>>>                    from drivers/thermal/samsung/exynos_tmu.c:14:
>>> drivers/thermal/samsung/exynos_tmu.c: In function 'exynos_tmu_update_bit':
>>> ./include/linux/cleanup.h:258:9: error: unknown type name
>>> 'class_exynos_tmu_data_t'
>>>     258 |         class_##_name##_t var
>>> __cleanup(class_##_name##_destructor) =   \
>>>         |         ^~~~~~
>>> ./include/linux/cleanup.h:309:9: note: in expansion of macro 'CLASS'
>>>     309 |         CLASS(_name, __UNIQUE_ID(guard))
>>>         |         ^~~~~
>>> drivers/thermal/samsung/exynos_tmu.c:338:9: note: in expansion of macro 'guard'
>>>     338 |         guard(exynos_tmu_data)(data);
>>>         |         ^~~~~
>>> drivers/thermal/samsung/exynos_tmu.c:338:9: error: cleanup argument
>>> not a function
>>
>> [snip]
>>
>> Right, you're missing the definition at the begging, like:
>>
>> DEFINE_GUARD(exynos_tmu_data, struct exynos_tmu_data *,
>> mutex_lock(&_T->lock),
>>                mutex_unlock(&_T->lock))
>>
>> below the struct exynos_tmu_data definition.
>>
>> Also, make sure you include the cleanup.h (it might not complain,
>> but it would be explicit and more clear)
> 
> Thanks for this tip.
> However, incorporating guard(exynos_tmu_data)(data); results
> in a recursive deadlock with the mutex during initialization, as this
> data structure is common to all the code configurations of Exynos TMU
> 

Fair enough, it would be just a cosmetic change, so do fight with it
too much.

Please continue in v4 with your former approach with the 'guard'.