Using of devm API leads to a certain order of releasing resources.
So all dependent resources which are not devm-wrapped should be deleted
with respect to devm-release order. Mutex is one of such objects that
often is bound to other resources and has no own devm wrapper.
Since mutex_destroy() actually does nothing in non-debug builds
frequently calling mutex_destroy() is just ignored which is safe for now
but wrong formally and can lead to a problem if mutex_destroy() is
extended so introduce devm_mutex_init().

Signed-off-by: George Stark <gnstark@salutedevices.com>
---
 include/linux/devm-helpers.h | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/include/linux/devm-helpers.h b/include/linux/devm-helpers.h
index 74891802200d..4043c3481d2e 100644
--- a/include/linux/devm-helpers.h
+++ b/include/linux/devm-helpers.h
@@ -24,6 +24,7 @@
  */
 
 #include <linux/device.h>
+#include <linux/mutex.h>
 #include <linux/workqueue.h>
 
 static inline void devm_delayed_work_drop(void *res)
@@ -76,4 +77,30 @@ static inline int devm_work_autocancel(struct device *dev,
 	return devm_add_action(dev, devm_work_drop, w);
 }
 
+#ifdef mutex_destroy
+static inline void devm_mutex_release(void *res)
+{
+	mutex_destroy(res);
+}
+#endif
+
+/**
+ * devm_mutex_init - Resource-managed mutex initialization
+ * @dev:	Device which lifetime mutex is bound to
+ * @lock:	Pointer to a mutex
+ *
+ * Initialize mutex which is automatically destroyed when the driver is detached.
+ *
+ * Returns: 0 on success or a negative error code on failure.
+ */
+static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
+{
+	mutex_init(lock);
+#ifdef mutex_destroy
+	return devm_add_action_or_reset(dev, devm_mutex_release, lock);
+#else
+	return 0;
+#endif
+}
+
 #endif
-- 
2.25.1

Le 13/12/2023 à 23:30, George Stark a écrit :
> [Vous ne recevez pas souvent de courriers de gnstark@salutedevices.com. Découvrez pourquoi ceci est important à https://aka.ms/LearnAboutSenderIdentification ]
> 
> Using of devm API leads to a certain order of releasing resources.
> So all dependent resources which are not devm-wrapped should be deleted
> with respect to devm-release order. Mutex is one of such objects that
> often is bound to other resources and has no own devm wrapper.
> Since mutex_destroy() actually does nothing in non-debug builds
> frequently calling mutex_destroy() is just ignored which is safe for now
> but wrong formally and can lead to a problem if mutex_destroy() is
> extended so introduce devm_mutex_init().

So you abandonned the idea of using mutex.h ?

I can't see the point to spread mutex functions into devm-helpers.h

Adding a mutex_destroy macro for this purpose looks odd. And if someone 
defines a new version of mutex_destroy() and forget the macro, it will 
go undetected.

Usually macros of that type serve the purpose of defining a fallback 
when the macro is not defined. In that case, when someone adds a new 
version without defining the macro, it gets detected because if 
conflicts with the fallback.
But in your case it works the other way round, so I will just go undetected.

For me the best solution remains to use mutex.h and have 
devm_mutex_init() defined or declared at the same place as mutex_destroy().


> 
> Signed-off-by: George Stark <gnstark@salutedevices.com>
> ---
>   include/linux/devm-helpers.h | 27 +++++++++++++++++++++++++++
>   1 file changed, 27 insertions(+)
> 
> diff --git a/include/linux/devm-helpers.h b/include/linux/devm-helpers.h
> index 74891802200d..4043c3481d2e 100644
> --- a/include/linux/devm-helpers.h
> +++ b/include/linux/devm-helpers.h
> @@ -24,6 +24,7 @@
>    */
> 
>   #include <linux/device.h>
> +#include <linux/mutex.h>
>   #include <linux/workqueue.h>
> 
>   static inline void devm_delayed_work_drop(void *res)
> @@ -76,4 +77,30 @@ static inline int devm_work_autocancel(struct device *dev,
>          return devm_add_action(dev, devm_work_drop, w);
>   }
> 
> +#ifdef mutex_destroy
> +static inline void devm_mutex_release(void *res)
> +{
> +       mutex_destroy(res);
> +}
> +#endif
> +
> +/**
> + * devm_mutex_init - Resource-managed mutex initialization
> + * @dev:       Device which lifetime mutex is bound to
> + * @lock:      Pointer to a mutex
> + *
> + * Initialize mutex which is automatically destroyed when the driver is detached.
> + *
> + * Returns: 0 on success or a negative error code on failure.
> + */
> +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
> +{
> +       mutex_init(lock);
> +#ifdef mutex_destroy
> +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
> +#else
> +       return 0;
> +#endif
> +}
> +
>   #endif
> --
> 2.25.1
> 

Hello Christophe

On 12/14/23 13:06, Christophe Leroy wrote:
> 
> 
...
> 
> So you abandonned the idea of using mutex.h ?

I'm not the one who make a choice here. The patch [1] you're talking 
about was seen by everyone but it seems like no one had shown interest.
For me personally approach with #define mutex_destroy is not very usual
but if even slight mixing device with mutex.h is unacceptable what else
can we do? Avoiding the need to allocate devm slot for empty 
mutex_destroy is more important.

Should I make series #4 with the patch [1] to give it a last chance?

[1] 
https://lore.kernel.org/lkml/377e4437-7051-4d88-ae68-1460bcd692e1@redhat.com/T/#m3f6df30ffccaccb1df4669a327f349164f572931


> I can't see the point to spread mutex functions into devm-helpers.h
> 
> Adding a mutex_destroy macro for this purpose looks odd. And if someone
> defines a new version of mutex_destroy() and forget the macro, it will
> go undetected.
> 
> Usually macros of that type serve the purpose of defining a fallback
> when the macro is not defined. In that case, when someone adds a new
> version without defining the macro, it gets detected because if
> conflicts with the fallback.
> But in your case it works the other way round, so I will just go undetected.
> 
> For me the best solution remains to use mutex.h and have
> devm_mutex_init() defined or declared at the same place as mutex_destroy().
> 


> 
>>
>> Signed-off-by: George Stark <gnstark@salutedevices.com>
>> ---
>>    include/linux/devm-helpers.h | 27 +++++++++++++++++++++++++++
>>    1 file changed, 27 insertions(+)
>>
>> diff --git a/include/linux/devm-helpers.h b/include/linux/devm-helpers.h
>> index 74891802200d..4043c3481d2e 100644
>> --- a/include/linux/devm-helpers.h
>> +++ b/include/linux/devm-helpers.h
>> @@ -24,6 +24,7 @@
>>     */
>>
>>    #include <linux/device.h>
>> +#include <linux/mutex.h>
>>    #include <linux/workqueue.h>
>>
>>    static inline void devm_delayed_work_drop(void *res)
>> @@ -76,4 +77,30 @@ static inline int devm_work_autocancel(struct device *dev,
>>           return devm_add_action(dev, devm_work_drop, w);
>>    }
>>
>> +#ifdef mutex_destroy
>> +static inline void devm_mutex_release(void *res)
>> +{
>> +       mutex_destroy(res);
>> +}
>> +#endif
>> +
>> +/**
>> + * devm_mutex_init - Resource-managed mutex initialization
>> + * @dev:       Device which lifetime mutex is bound to
>> + * @lock:      Pointer to a mutex
>> + *
>> + * Initialize mutex which is automatically destroyed when the driver is detached.
>> + *
>> + * Returns: 0 on success or a negative error code on failure.
>> + */
>> +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
>> +{
>> +       mutex_init(lock);
>> +#ifdef mutex_destroy
>> +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
>> +#else
>> +       return 0;
>> +#endif
>> +}
>> +
>>    #endif
>> --
>> 2.25.1
>>

-- 
Best regards
George

Le 14/12/2023 à 13:48, George Stark a écrit :
> [Vous ne recevez pas souvent de courriers de gnstark@salutedevices.com. 
> Découvrez pourquoi ceci est important à 
> https://aka.ms/LearnAboutSenderIdentification ]
> 
> Hello Christophe
> 
> On 12/14/23 13:06, Christophe Leroy wrote:
>>
>>
> ...
>>
>> So you abandonned the idea of using mutex.h ?
> 
> I'm not the one who make a choice here. The patch [1] you're talking
> about was seen by everyone but it seems like no one had shown interest.
> For me personally approach with #define mutex_destroy is not very usual
> but if even slight mixing device with mutex.h is unacceptable what else
> can we do? Avoiding the need to allocate devm slot for empty
> mutex_destroy is more important.
> 

Why would a forward declaration of struct device in mutex.h be 
unacceptable when it is done in so many headers ?

$ git grep "struct device;" include/ | wc -l
164



> Should I make series #4 with the patch [1] to give it a last chance?

Yes, lets give it a try

> 
> [1]
> https://lore.kernel.org/lkml/377e4437-7051-4d88-ae68-1460bcd692e1@redhat.com/T/#m3f6df30ffccaccb1df4669a327f349164f572931
> 

Christophe

On Thu, Dec 14, 2023 at 3:00 PM Christophe Leroy
<christophe.leroy@csgroup.eu> wrote:
> Le 14/12/2023 à 13:48, George Stark a écrit :
> > [Vous ne recevez pas souvent de courriers de gnstark@salutedevices.com.
> > Découvrez pourquoi ceci est important à
> > https://aka.ms/LearnAboutSenderIdentification ]
> > On 12/14/23 13:06, Christophe Leroy wrote:

...

> >> So you abandonned the idea of using mutex.h ?
> >
> > I'm not the one who make a choice here. The patch [1] you're talking
> > about was seen by everyone but it seems like no one had shown interest.
> > For me personally approach with #define mutex_destroy is not very usual
> > but if even slight mixing device with mutex.h is unacceptable what else
> > can we do? Avoiding the need to allocate devm slot for empty
> > mutex_destroy is more important.
> >
>
> Why would a forward declaration of struct device in mutex.h be
> unacceptable when it is done in so many headers ?
>
> $ git grep "struct device;" include/ | wc -l
> 164

I believe the main misunderstanding here is where to put the
implementation. AFAIU Christophe wants the implementation to be in the
very same _C_-file where mutex_destroy() is defined. mutex.h in this
case indeed requires the only forward declaration and hence doesn't
need to include device.h.

-- 
With Best Regards,
Andy Shevchenko

Hi,

On 12/14/23 11:06, Christophe Leroy wrote:
> 
> 
> Le 13/12/2023 à 23:30, George Stark a écrit :
>> [Vous ne recevez pas souvent de courriers de gnstark@salutedevices.com. Découvrez pourquoi ceci est important à https://aka.ms/LearnAboutSenderIdentification ]
>>
>> Using of devm API leads to a certain order of releasing resources.
>> So all dependent resources which are not devm-wrapped should be deleted
>> with respect to devm-release order. Mutex is one of such objects that
>> often is bound to other resources and has no own devm wrapper.
>> Since mutex_destroy() actually does nothing in non-debug builds
>> frequently calling mutex_destroy() is just ignored which is safe for now
>> but wrong formally and can lead to a problem if mutex_destroy() is
>> extended so introduce devm_mutex_init().
> 
> So you abandonned the idea of using mutex.h ?
> 
> I can't see the point to spread mutex functions into devm-helpers.h
> 
> Adding a mutex_destroy macro for this purpose looks odd. And if someone 
> defines a new version of mutex_destroy() and forget the macro, it will 
> go undetected.
> 
> Usually macros of that type serve the purpose of defining a fallback 
> when the macro is not defined. In that case, when someone adds a new 
> version without defining the macro, it gets detected because if 
> conflicts with the fallback.
> But in your case it works the other way round, so I will just go undetected.
> 
> For me the best solution remains to use mutex.h and have 
> devm_mutex_init() defined or declared at the same place as mutex_destroy().

FWIW defining devm_mutex_init() in mutex.h is fine
with me and makes sense to me. I also agree that putting
it there would be better if that is acceptable for
the mutex maintainers.

devm-helpers.h is there for helpers which don't fit
in another place.

Regards,

Hans




> 
> 
>>
>> Signed-off-by: George Stark <gnstark@salutedevices.com>
>> ---
>>   include/linux/devm-helpers.h | 27 +++++++++++++++++++++++++++
>>   1 file changed, 27 insertions(+)
>>
>> diff --git a/include/linux/devm-helpers.h b/include/linux/devm-helpers.h
>> index 74891802200d..4043c3481d2e 100644
>> --- a/include/linux/devm-helpers.h
>> +++ b/include/linux/devm-helpers.h
>> @@ -24,6 +24,7 @@
>>    */
>>
>>   #include <linux/device.h>
>> +#include <linux/mutex.h>
>>   #include <linux/workqueue.h>
>>
>>   static inline void devm_delayed_work_drop(void *res)
>> @@ -76,4 +77,30 @@ static inline int devm_work_autocancel(struct device *dev,
>>          return devm_add_action(dev, devm_work_drop, w);
>>   }
>>
>> +#ifdef mutex_destroy
>> +static inline void devm_mutex_release(void *res)
>> +{
>> +       mutex_destroy(res);
>> +}
>> +#endif
>> +
>> +/**
>> + * devm_mutex_init - Resource-managed mutex initialization
>> + * @dev:       Device which lifetime mutex is bound to
>> + * @lock:      Pointer to a mutex
>> + *
>> + * Initialize mutex which is automatically destroyed when the driver is detached.
>> + *
>> + * Returns: 0 on success or a negative error code on failure.
>> + */
>> +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
>> +{
>> +       mutex_init(lock);
>> +#ifdef mutex_destroy
>> +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
>> +#else
>> +       return 0;
>> +#endif
>> +}
>> +
>>   #endif
>> --
>> 2.25.1
>>

On Thu, Dec 14, 2023 at 12:30 AM George Stark <gnstark@salutedevices.com> wrote:
>
> Using of devm API leads to a certain order of releasing resources.
> So all dependent resources which are not devm-wrapped should be deleted
> with respect to devm-release order. Mutex is one of such objects that
> often is bound to other resources and has no own devm wrapper.
> Since mutex_destroy() actually does nothing in non-debug builds
> frequently calling mutex_destroy() is just ignored which is safe for now
> but wrong formally and can lead to a problem if mutex_destroy() is
> extended so introduce devm_mutex_init().

...

> +#ifdef mutex_destroy
> +static inline void devm_mutex_release(void *res)
> +{
> +       mutex_destroy(res);
> +}
> +#endif
> +
> +/**
> + * devm_mutex_init - Resource-managed mutex initialization
> + * @dev:       Device which lifetime mutex is bound to
> + * @lock:      Pointer to a mutex
> + *
> + * Initialize mutex which is automatically destroyed when the driver is detached.
> + *
> + * Returns: 0 on success or a negative error code on failure.
> + */
> +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
> +{
> +       mutex_init(lock);
> +#ifdef mutex_destroy
> +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
> +#else
> +       return 0;
> +#endif
> +}

If this is going to be accepted, you may decrease the amount of ifdeffery.

#ifdef ...
#else
#define devm_mutex_init(dev, lock)  mutex_init(lock)
#endif


-- 
With Best Regards,
Andy Shevchenko

On Thu, Dec 14, 2023 at 12:36 AM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:
> On Thu, Dec 14, 2023 at 12:30 AM George Stark <gnstark@salutedevices.com> wrote:
> >
> > Using of devm API leads to a certain order of releasing resources.
> > So all dependent resources which are not devm-wrapped should be deleted
> > with respect to devm-release order. Mutex is one of such objects that
> > often is bound to other resources and has no own devm wrapper.
> > Since mutex_destroy() actually does nothing in non-debug builds
> > frequently calling mutex_destroy() is just ignored which is safe for now
> > but wrong formally and can lead to a problem if mutex_destroy() is
> > extended so introduce devm_mutex_init().

...

> > +#ifdef mutex_destroy
> > +static inline void devm_mutex_release(void *res)
> > +{
> > +       mutex_destroy(res);
> > +}
> > +#endif
> > +
> > +/**
> > + * devm_mutex_init - Resource-managed mutex initialization
> > + * @dev:       Device which lifetime mutex is bound to
> > + * @lock:      Pointer to a mutex
> > + *
> > + * Initialize mutex which is automatically destroyed when the driver is detached.
> > + *
> > + * Returns: 0 on success or a negative error code on failure.
> > + */
> > +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
> > +{
> > +       mutex_init(lock);
> > +#ifdef mutex_destroy
> > +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
> > +#else
> > +       return 0;
> > +#endif
> > +}
>
> If this is going to be accepted, you may decrease the amount of ifdeffery.
>
> #ifdef ...
> #else
> #define devm_mutex_init(dev, lock)  mutex_init(lock)

More precisely ({ mutex_init(lock); 0; }) or as a static inline...

> #endif

-- 
With Best Regards,
Andy Shevchenko

Hi,

On 12/13/23 23:38, Andy Shevchenko wrote:
> On Thu, Dec 14, 2023 at 12:36 AM Andy Shevchenko
> <andy.shevchenko@gmail.com> wrote:
>> On Thu, Dec 14, 2023 at 12:30 AM George Stark <gnstark@salutedevices.com> wrote:
>>>
>>> Using of devm API leads to a certain order of releasing resources.
>>> So all dependent resources which are not devm-wrapped should be deleted
>>> with respect to devm-release order. Mutex is one of such objects that
>>> often is bound to other resources and has no own devm wrapper.
>>> Since mutex_destroy() actually does nothing in non-debug builds
>>> frequently calling mutex_destroy() is just ignored which is safe for now
>>> but wrong formally and can lead to a problem if mutex_destroy() is
>>> extended so introduce devm_mutex_init().
> 
> ...
> 
>>> +#ifdef mutex_destroy
>>> +static inline void devm_mutex_release(void *res)
>>> +{
>>> +       mutex_destroy(res);
>>> +}
>>> +#endif
>>> +
>>> +/**
>>> + * devm_mutex_init - Resource-managed mutex initialization
>>> + * @dev:       Device which lifetime mutex is bound to
>>> + * @lock:      Pointer to a mutex
>>> + *
>>> + * Initialize mutex which is automatically destroyed when the driver is detached.
>>> + *
>>> + * Returns: 0 on success or a negative error code on failure.
>>> + */
>>> +static inline int devm_mutex_init(struct device *dev, struct mutex *lock)
>>> +{
>>> +       mutex_init(lock);
>>> +#ifdef mutex_destroy
>>> +       return devm_add_action_or_reset(dev, devm_mutex_release, lock);
>>> +#else
>>> +       return 0;
>>> +#endif
>>> +}
>>
>> If this is going to be accepted, you may decrease the amount of ifdeffery.
>>
>> #ifdef ...
>> #else
>> #define devm_mutex_init(dev, lock)  mutex_init(lock)
> 
> More precisely ({ mutex_init(lock); 0; }) or as a static inline...

With a static inline we are pretty much back to the original
v3 patch.

I believe the best way to reduce the ifdef-ery is to remove
the #ifdef around devm_mutex_release() having unused
static inline ... functions in .h files is quite common,
so this one does not need a #ifdef around it and with
that removed we are down to just one #ifdef so just
removing the #ifdef around devm_mutex_release() seems
the best fix.

With that fixed you may add my:

Reviewed-by: Hans de Goede <hdegoede@redhat.com>

to the patch and I'm fine with this being routed
upstream through whatever tree is convenient.

Regards,

Hans