Skip to contentSkip to conversation

powerpc: kernel: Change the order of of_node_put()

PATCH In review Liang He 2 participants 8 messages Jun 17, 2022 – Jun 20, 2022 PowerPC

lore
Share & tools
lore

Liang He

author
#0 lore raw

PATCH 1/1 powerpc: kernel: Change the order of of_node_put()

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

---
 arch/powerpc/kernel/setup-common.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/kernel/setup-common.c +1 -1
index eb0077b302e2..761817d1f4db 100644
@@ -563,9 +563,9 @@ static __init int add_pcspkr(void)
 	int ret;
 
 	np = of_find_compatible_node(NULL, NULL, "pnpPNP,100");
-	of_node_put(np);
 	if (!np)
 		return -ENODEV;
+	of_node_put(np);
 
 	pd = platform_device_alloc("pcspkr", -1);
 	if (!pd)
Signature · 1 line
2.25.1
#1 lore raw

Le 17/06/2022 à 13:26, Liang He a écrit :

Quoted 2 lines

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

Why is it better ?

/**
* of_node_put() - Decrement refcount of a node
* @node: Node to dec refcount, NULL is supported to simplify writing of
* callers
*/
void of_node_put(struct device_node *node)

{
	if (node)
		kobject_put(&node->kobj);
}

EXPORT_SYMBOL(of_node_put);

Christophe

Quoted 21 lines
---
  arch/powerpc/kernel/setup-common.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/kernel/setup-common.c +1 -1
index eb0077b302e2..761817d1f4db 100644
@@ -563,9 +563,9 @@ static __init int add_pcspkr(void)
  	int ret;
 
  	np = of_find_compatible_node(NULL, NULL, "pnpPNP,100");
-	of_node_put(np);
  	if (!np)
  		return -ENODEV;
+	of_node_put(np);
 
  	pd = platform_device_alloc("pcspkr", -1);
  	if (!pd)
author
#2 lore raw
Quoted 20 lines · 在 2022-06-18 15:13:13,"Christophe Leroy" <[email protected]> 写道:

Le 17/06/2022 à 13:26, Liang He a écrit :

Quoted 2 lines

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

Why is it better ?

/**
* of_node_put() - Decrement refcount of a node
* @node: Node to dec refcount, NULL is supported to simplify writing of
* callers
*/
void of_node_put(struct device_node *node)

{
	if (node)
		kobject_put(&node->kobj);
}
Quoted 5 lines

EXPORT_SYMBOL(of_node_put);

Christophe

Hi, Christophe.

Thanks for your reply and I want to have a discussion.

In my thought, xxx_put(pointer)'s semantic usually means this reference has been used done and will not be used anymore. Is this semantic more reasonable, right?

Besides, if the np is NULL, we can just return and save a cpu time for the xxx_put() call.

Otherwise, I prefer to call it 'use(check)-after-put'.

In fact, I have meet many other 'use(check)-after-put' instances after I send this patch-commit, so I am waiting for this discussion.

This is just my thought, it may be wrong.

Anyway, thanks for your reply.

Liang

Quoted 23 lines
Quoted 23 lines
---
  arch/powerpc/kernel/setup-common.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/kernel/setup-common.c +1 -1
index eb0077b302e2..761817d1f4db 100644
@@ -563,9 +563,9 @@ static __init int add_pcspkr(void)
  	int ret;
 
  	np = of_find_compatible_node(NULL, NULL, "pnpPNP,100");
-	of_node_put(np);
  	if (!np)
  		return -ENODEV;
+	of_node_put(np);
 
  	pd = platform_device_alloc("pcspkr", -1);
  	if (!pd)
#3 lore raw

Le 18/06/2022 à 10:03, Liang He a écrit :

Quoted 45 lines · 在 2022-06-18 15:13:13,"Christophe Leroy" <[email protected]> 写道:
Quoted 20 lines

Le 17/06/2022 à 13:26, Liang He a écrit :

Quoted 2 lines

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

Why is it better ?

/**
* of_node_put() - Decrement refcount of a node
* @node: Node to dec refcount, NULL is supported to simplify writing of
* callers
*/
void of_node_put(struct device_node *node)

{
	if (node)
		kobject_put(&node->kobj);
}
Quoted 5 lines

EXPORT_SYMBOL(of_node_put);

Christophe

Hi, Christophe.

Thanks for your reply and I want to have a discussion.

In my thought, xxx_put(pointer)'s semantic usually means this reference has been used done and will not be used anymore. Is this semantic more reasonable, right?

Besides, if the np is NULL, we can just return and save a cpu time for the xxx_put() call.

Otherwise, I prefer to call it 'use(check)-after-put'.

Quoted 7 lines

In fact, I have meet many other 'use(check)-after-put' instances after I send this patch-commit, so I am waiting for this discussion.

This is just my thought, it may be wrong.

Anyway, thanks for your reply.

Well in principle you are right, in an ideal world it should be like that. However, you have to wonder if it is worth the churn. The CPU cycle argument is valid only if that function is used in a hot path. But as we are talking about error handling, it can't be a hot path.

Taking into account the comment associated of of_node_put : "NULL is supported to simplify writing of callers", it means that usage is valid, just like it is with function kfree() after a kmalloc().

So in a new developpement, or when doing real modifications to a driver, that kind of change can be done ideally. However for drivers that have been there for years without any change, ask yourself if it is worth the churn. You spend time on it, you require other people to spend time on it for reviewing and applying your patches and during that time they don't do other things that could have been more usefull.

So unless this change is part of a more global patch, I think it is not worth the effort.

By the way, also for all your other patches, I think you should start doing all the changes locally on your side, and when you are finished try to group things together in bigger patches per area instead of sending one by one. I see you have already started doing that for opal/powernv for instance, but there are still individual powernv/opal in the queue. I think you should group all together in a single patch. And same for other areas, please try to minimise the number of patches. We don't link huge bombs that modify all the kernel at once, but you can group things together, one patch for powerpc core parts, one patch for each platform in arch/powerpc/platforms/ etc ...

Christophe

author
#4 lore raw
Quoted 52 lines · Christophe Leroy

Le 18/06/2022 à 10:03, Liang He a écrit :

Quoted 49 lines · 在 2022-06-18 15:13:13,"Christophe Leroy" <[email protected]> 写道:
Quoted 20 lines

Le 17/06/2022 à 13:26, Liang He a écrit :

Quoted 2 lines

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

Why is it better ?

/**
* of_node_put() - Decrement refcount of a node
* @node: Node to dec refcount, NULL is supported to simplify writing of
* callers
*/
void of_node_put(struct device_node *node)

{
	if (node)
		kobject_put(&node->kobj);
}
Quoted 5 lines

EXPORT_SYMBOL(of_node_put);

Christophe

Hi, Christophe.

Thanks for your reply and I want to have a discussion.

In my thought, xxx_put(pointer)'s semantic usually means this reference has been used done and will not be used anymore. Is this semantic more reasonable, right?

Besides, if the np is NULL, we can just return and save a cpu time for the xxx_put() call.

Otherwise, I prefer to call it 'use(check)-after-put'.

In fact, I have meet many other 'use(check)-after-put' instances after I send this patch-commit, so I am waiting for this discussion.

Quoted 9 lines
Quoted 3 lines

This is just my thought, it may be wrong.

Anyway, thanks for your reply.

Well in principle you are right, in an ideal world it should be like that. However, you have to wonder if it is worth the churn. The CPU cycle argument is valid only if that function is used in a hot path. But as we are talking about error handling, it can't be a hot path.

Thanks very much for this valuable lesson.

Quoted 11 lines

Taking into account the comment associated of of_node_put : "NULL is supported to simplify writing of callers", it means that usage is valid, just like it is with function kfree() after a kmalloc().

So in a new developpement, or when doing real modifications to a driver, that kind of change can be done ideally. However for drivers that have been there for years without any change, ask yourself if it is worth the churn. You spend time on it, you require other people to spend time on it for reviewing and applying your patches and during that time they don't do other things that could have been more usefull.

Thanks for you advice, I will keep it in my mind before I send a new patch.

Quoted 14 lines

So unless this change is part of a more global patch, I think it is not worth the effort.

By the way, also for all your other patches, I think you should start doing all the changes locally on your side, and when you are finished try to group things together in bigger patches per area instead of sending one by one. I see you have already started doing that for opal/powernv for instance, but there are still individual powernv/opal in the queue. I think you should group all together in a single patch. And same for other areas, please try to minimise the number of patches. We don't link huge bombs that modify all the kernel at once, but you can group things together, one patch for powerpc core parts, one patch for each platform in arch/powerpc/platforms/ etc ...

You are right and I will follow this principle in future patching work. While It is too exciting for me to begin the patching work , I should have grouped my patches.

Quoted 2 lines

Christophe

Thanks again, Christophe.

Liang

author
#5 lore raw
In reply to Christophe Leroy
Quoted 87 lines · Christophe Leroy

Le 18/06/2022 à 10:03, Liang He a écrit :

Quoted 45 lines · 在 2022-06-18 15:13:13,"Christophe Leroy" <[email protected]> 写道:
Quoted 20 lines

Le 17/06/2022 à 13:26, Liang He a écrit :

Quoted 2 lines

In add_pcspkr(), it is better to call of_node_put() after the 'if(!np)' check.

Why is it better ?

/**
* of_node_put() - Decrement refcount of a node
* @node: Node to dec refcount, NULL is supported to simplify writing of
* callers
*/
void of_node_put(struct device_node *node)

{
	if (node)
		kobject_put(&node->kobj);
}
Quoted 5 lines

EXPORT_SYMBOL(of_node_put);

Christophe

Hi, Christophe.

Thanks for your reply and I want to have a discussion.

In my thought, xxx_put(pointer)'s semantic usually means this reference has been used done and will not be used anymore. Is this semantic more reasonable, right?

Besides, if the np is NULL, we can just return and save a cpu time for the xxx_put() call.

Otherwise, I prefer to call it 'use(check)-after-put'.

Quoted 7 lines

In fact, I have meet many other 'use(check)-after-put' instances after I send this patch-commit, so I am waiting for this discussion.

This is just my thought, it may be wrong.

Anyway, thanks for your reply.

Well in principle you are right, in an ideal world it should be like that. However, you have to wonder if it is worth the churn. The CPU cycle argument is valid only if that function is used in a hot path. But as we are talking about error handling, it can't be a hot path.

Taking into account the comment associated of of_node_put : "NULL is supported to simplify writing of callers", it means that usage is valid, just like it is with function kfree() after a kmalloc().

So in a new developpement, or when doing real modifications to a driver, that kind of change can be done ideally. However for drivers that have been there for years without any change, ask yourself if it is worth the churn. You spend time on it, you require other people to spend time on it for reviewing and applying your patches and during that time they don't do other things that could have been more usefull.

So unless this change is part of a more global patch, I think it is not worth the effort.

By the way, also for all your other patches, I think you should start doing all the changes locally on your side, and when you are finished try to group things together in bigger patches per area instead of sending one by one. I see you have already started doing that for opal/powernv for instance, but there are still individual powernv/opal in the queue. I think you should group all together in a single patch. And same for other areas, please try to minimise the number of patches. We don't link huge bombs that modify all the kernel at once, but you can group things together, one patch for powerpc core parts, one patch for each platform in arch/powerpc/platforms/ etc ...

Quoted 1 line

Christophe

Hi, Christophe.

Sorry to trobule you again.

Now I have found other bugs in same directories (i.e., arch/powerpc/sysdev), with the ones I have sent but not recieved acked-by or confirmed email.

So I need to merge the old ones into the new ones as a PATCH-v2 and then resend the old ones ?
or just use a new PATCH to send only new ones?

I am afraid to make new trouble for maintainers, so can you share your valuable experience?

Thanks very much.

Liang

#6 lore raw

Hi,

Le 20/06/2022 à 11:23, Liang He a écrit :

Quoted 8 lines

Hi, Christophe.

Sorry to trobule you again.

Now I have found other bugs in same directories (i.e., arch/powerpc/sysdev), with the ones I have sent but not recieved acked-by or confirmed email.

Quoted 7 lines

So I need to merge the old ones into the new ones as a PATCH-v2 and then resend the old ones ?
or just use a new PATCH to send only new ones?

I am afraid to make new trouble for maintainers, so can you share your valuable experience?

Here is the list of your patches :
https://patchwork.ozlabs.org/project/linuxppc-dev/list/?submitter=84258

From my point of view, for all the patches that are still in status
"new" it is better than you send a v2 with more things into a single patch. When the patch is in "under review" state, it is better to not update it anymore.

So in the list there are for instance several patches for powernv, so it would be good if you can regroup all of them in a single v2 patch.

Christophe

author
#7 lore raw
Quoted 27 lines · Christophe Leroy

Hi,

Le 20/06/2022 à 11:23, Liang He a écrit :

Quoted 8 lines

Hi, Christophe.

Sorry to trobule you again.

Now I have found other bugs in same directories (i.e., arch/powerpc/sysdev), with the ones I have sent but not recieved acked-by or confirmed email.

Quoted 7 lines

So I need to merge the old ones into the new ones as a PATCH-v2 and then resend the old ones ?
or just use a new PATCH to send only new ones?

I am afraid to make new trouble for maintainers, so can you share your valuable experience?

Here is the list of your patches :
https://patchwork.ozlabs.org/project/linuxppc-dev/list/?submitter=84258

From my point of view, for all the patches that are still in status "new" it is better than you send a v2 with more things into a single patch. When the patch is in "under review" state, it is better to not update it anymore.

Quoted 4 lines

So in the list there are for instance several patches for powernv, so it would be good if you can regroup all of them in a single v2 patch.

Christophe

Thanks, Christophe.

I will follow your rules and try to group the 'new' state ones.