[PATCH] mm: memory_hotplug: use put_device() if device_register fail

[PATCH] mm: memory_hotplug: use put_device() if device_register fail

[Arvind Yadav]

if device_register() returned an error. Always use put_device()
to give up the initialized reference and release allocated memory.

Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
---
 drivers/base/memory.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/base/memory.c b/drivers/base/memory.c
index bffe861..f5e5601 100644
--- a/drivers/base/memory.c
+++ b/drivers/base/memory.c
@@ -649,13 +649,19 @@ static const struct attribute_group *memory_memblk_attr_groups[] = {
 static
 int register_memory(struct memory_block *memory)
 {
+	int ret;
+
 	memory->dev.bus = &memory_subsys;
 	memory->dev.id = memory->start_section_nr / sections_per_block;
 	memory->dev.release = memory_block_release;
 	memory->dev.groups = memory_memblk_attr_groups;
 	memory->dev.offline = memory->state == MEM_OFFLINE;
 
-	return device_register(&memory->dev);
+	ret = device_register(&memory->dev);
+	if (ret)
+		put_device(&memory->dev);
+
+	return ret;
 }
 
 static int init_memory_block(struct memory_block **memory,
-- 
2.7.4

Re: [PATCH] mm: memory_hotplug: use put_device() if device_register fail

[Michal Hocko]

On Thu 26-04-18 21:12:09, Arvind Yadav wrote:
> if device_register() returned an error. Always use put_device()
> to give up the initialized reference and release allocated memory.

Is this patch correct? The docummentation says
 * NOTE: _Never_ directly free @dev after calling this function, even
 * if it returned an error! Always use put_device() to give up your
 * reference instead.

but we do not have _our_ reference in this path AFAICS. Maybe this is
just a documentation issue? How have you tested this change btw.?

> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
> ---
>  drivers/base/memory.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> index bffe861..f5e5601 100644
> --- a/drivers/base/memory.c
> +++ b/drivers/base/memory.c
> @@ -649,13 +649,19 @@ static const struct attribute_group *memory_memblk_attr_groups[] = {
>  static
>  int register_memory(struct memory_block *memory)
>  {
> +	int ret;
> +
>  	memory->dev.bus = &memory_subsys;
>  	memory->dev.id = memory->start_section_nr / sections_per_block;
>  	memory->dev.release = memory_block_release;
>  	memory->dev.groups = memory_memblk_attr_groups;
>  	memory->dev.offline = memory->state == MEM_OFFLINE;
>  
> -	return device_register(&memory->dev);
> +	ret = device_register(&memory->dev);
> +	if (ret)
> +		put_device(&memory->dev);
> +
> +	return ret;
>  }
>  
>  static int init_memory_block(struct memory_block **memory,
> -- 
> 2.7.4

-- 
Michal Hocko
SUSE Labs

Re: [PATCH] mm: memory_hotplug: use put_device() if device_register fail

[arvindY]

On Friday 27 April 2018 08:26 PM, Michal Hocko wrote:
> On Thu 26-04-18 21:12:09, Arvind Yadav wrote:
>> if device_register() returned an error. Always use put_device()
>> to give up the initialized reference and release allocated memory.
> Is this patch correct? The docummentation says
>   * NOTE: _Never_ directly free @dev after calling this function, even
>   * if it returned an error! Always use put_device() to give up your
>   * reference instead.
>
> but we do not have _our_ reference in this path AFAICS. Maybe this is
> just a documentation issue? How have you tested this change btw.?
The document is correct. Here device_register() will initialize object by
making reference count as 1 and also increment reference count for device.

device_register() {
    device_initialize()->kobject_init()->kref_init() - initialize 
object( reference count = 1).
    device_add()->get_device() - increment reference count for device.
}

If device_register() will fail then we have to release the object by making
reference count 0. So we need to call put_object() which will release the
object and other resources like memory etc. so long as the reference
count is nonzero, the object continue to exist in memory.

I have not tested this peace of code. But I have tested other code which
is using Kboject.

~arvind
>
>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
>> ---
>>   drivers/base/memory.c | 8 +++++++-
>>   1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
>> index bffe861..f5e5601 100644
>> --- a/drivers/base/memory.c
>> +++ b/drivers/base/memory.c
>> @@ -649,13 +649,19 @@ static const struct attribute_group *memory_memblk_attr_groups[] = {
>>   static
>>   int register_memory(struct memory_block *memory)
>>   {
>> +	int ret;
>> +
>>   	memory->dev.bus = &memory_subsys;
>>   	memory->dev.id = memory->start_section_nr / sections_per_block;
>>   	memory->dev.release = memory_block_release;
>>   	memory->dev.groups = memory_memblk_attr_groups;
>>   	memory->dev.offline = memory->state == MEM_OFFLINE;
>>   
>> -	return device_register(&memory->dev);
>> +	ret = device_register(&memory->dev);
>> +	if (ret)
>> +		put_device(&memory->dev);
>> +
>> +	return ret;
>>   }
>>   
>>   static int init_memory_block(struct memory_block **memory,
>> -- 
>> 2.7.4

Re: [PATCH] mm: memory_hotplug: use put_device() if device_register fail

[Michal Hocko]

On Sat 28-04-18 11:05:44, arvindY wrote:
> 
> 
> On Friday 27 April 2018 08:26 PM, Michal Hocko wrote:
> > On Thu 26-04-18 21:12:09, Arvind Yadav wrote:
> > > if device_register() returned an error. Always use put_device()
> > > to give up the initialized reference and release allocated memory.
> > Is this patch correct? The docummentation says
> >   * NOTE: _Never_ directly free @dev after calling this function, even
> >   * if it returned an error! Always use put_device() to give up your
> >   * reference instead.
> > 
> > but we do not have _our_ reference in this path AFAICS. Maybe this is
> > just a documentation issue? How have you tested this change btw.?
> The document is correct. Here device_register() will initialize object by
> making reference count as 1 and also increment reference count for device.
> 
> device_register() {
>    device_initialize()->kobject_init()->kref_init() - initialize object(
> reference count = 1).
>    device_add()->get_device() - increment reference count for device.
> }
> 
> If device_register() will fail then we have to release the object by making
> reference count 0.

I am confused. If device_register intializes the reference to 1 then why
it doesn't decrement it on a failure path and rather expects the caller
to do that? The doc as I read it means that we should only drop a
reference if we have our own one before calling device_register.

Or what do I miss here?
-- 
Michal Hocko
SUSE Labs