[PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[Chandra S Gorentla]

The spin_lock_irqsave is moved to just beginning of critical section.
This change moves a couple of return statements out of the lock.

Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
---
 drivers/staging/wilc1000/wilc_msgqueue.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
index d5ebd6d..284a3f5 100644
--- a/drivers/staging/wilc1000/wilc_msgqueue.c
+++ b/drivers/staging/wilc1000/wilc_msgqueue.c
@@ -72,8 +72,6 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
 		goto ERRORHANDLER;
 	}
 
-	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
-
 	/* construct a new message */
 	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
 	if (!pstrMessage)
@@ -87,6 +85,8 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
 	}
 	memcpy(pstrMessage->pvBuffer, pvSendBuffer, u32SendBufferSize);
 
+	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
+
 	/* add it to the message queue */
 	if (!pHandle->pstrMessageList) {
 		pHandle->pstrMessageList  = pstrMessage;
-- 
2.1.4

[PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Chandra S Gorentla]

 - kfree is being called for the members of the queue without
   de-queuing them; they are just inserted within this function;
   they are supposed to be de-queued and freed in a function
   for receiving the queue items
 - goto statements are removed
 - After kfree correction, there is no need for target block
   of goto statement; hence it is removed

Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
---
 drivers/staging/wilc1000/wilc_msgqueue.c | 22 ++++++----------------
 1 file changed, 6 insertions(+), 16 deletions(-)

diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
index 284a3f5..eae90be 100644
--- a/drivers/staging/wilc1000/wilc_msgqueue.c
+++ b/drivers/staging/wilc1000/wilc_msgqueue.c
@@ -56,32 +56,30 @@ int wilc_mq_destroy(WILC_MsgQueueHandle *pHandle)
 int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
 			     const void *pvSendBuffer, u32 u32SendBufferSize)
 {
-	int result = 0;
 	unsigned long flags;
 	Message *pstrMessage = NULL;
 
 	if ((!pHandle) || (u32SendBufferSize == 0) || (!pvSendBuffer)) {
 		PRINT_ER("pHandle or pvSendBuffer is null\n");
-		result = -EFAULT;
-		goto ERRORHANDLER;
+		return -EFAULT;
 	}
 
 	if (pHandle->bExiting) {
 		PRINT_ER("pHandle fail\n");
-		result = -EFAULT;
-		goto ERRORHANDLER;
+		return -EFAULT;
 	}
 
 	/* construct a new message */
 	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
 	if (!pstrMessage)
 		return -ENOMEM;
+
 	pstrMessage->u32Length = u32SendBufferSize;
 	pstrMessage->pstrNext = NULL;
 	pstrMessage->pvBuffer = kmalloc(u32SendBufferSize, GFP_ATOMIC);
 	if (!pstrMessage->pvBuffer) {
-		result = -ENOMEM;
-		goto ERRORHANDLER;
+		kfree(pstrMessage);
+		return -ENOMEM;
 	}
 	memcpy(pstrMessage->pvBuffer, pvSendBuffer, u32SendBufferSize);
 
@@ -102,15 +100,7 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
 	spin_unlock_irqrestore(&pHandle->strCriticalSection, flags);
 
 	up(&pHandle->hSem);
-
-ERRORHANDLER:
-	/* error occured, free any allocations */
-	if (pstrMessage) {
-		kfree(pstrMessage->pvBuffer);
-		kfree(pstrMessage);
-	}
-
-	return result;
+	return 0;
 }
 
 /*!
-- 
2.1.4

Re: [PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[Greg KH]

On Sat, Oct 03, 2015 at 02:57:29PM +0530, Chandra S Gorentla wrote:
> The spin_lock_irqsave is moved to just beginning of critical section.
> This change moves a couple of return statements out of the lock.
> 
> Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
> ---
>  drivers/staging/wilc1000/wilc_msgqueue.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
> index d5ebd6d..284a3f5 100644
> --- a/drivers/staging/wilc1000/wilc_msgqueue.c
> +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
> @@ -72,8 +72,6 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>  		goto ERRORHANDLER;
>  	}
>  
> -	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
> -
>  	/* construct a new message */
>  	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);

As you have moved the lock, can you also change this to GFP_KERNEL as
well because we do not have a lock held?

And how have you tested that this is ok?  What is this lock trying to
protect?

thanks,

greg k-h

Re: [PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Greg KH]

On Sat, Oct 03, 2015 at 02:57:30PM +0530, Chandra S Gorentla wrote:
>  - kfree is being called for the members of the queue without
>    de-queuing them; they are just inserted within this function;
>    they are supposed to be de-queued and freed in a function
>    for receiving the queue items
>  - goto statements are removed
>  - After kfree correction, there is no need for target block
>    of goto statement; hence it is removed
> 
> Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
> ---
>  drivers/staging/wilc1000/wilc_msgqueue.c | 22 ++++++----------------
>  1 file changed, 6 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
> index 284a3f5..eae90be 100644
> --- a/drivers/staging/wilc1000/wilc_msgqueue.c
> +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
> @@ -56,32 +56,30 @@ int wilc_mq_destroy(WILC_MsgQueueHandle *pHandle)
>  int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>  			     const void *pvSendBuffer, u32 u32SendBufferSize)
>  {
> -	int result = 0;
>  	unsigned long flags;
>  	Message *pstrMessage = NULL;
>  
>  	if ((!pHandle) || (u32SendBufferSize == 0) || (!pvSendBuffer)) {
>  		PRINT_ER("pHandle or pvSendBuffer is null\n");
> -		result = -EFAULT;
> -		goto ERRORHANDLER;
> +		return -EFAULT;
>  	}
>  
>  	if (pHandle->bExiting) {
>  		PRINT_ER("pHandle fail\n");
> -		result = -EFAULT;
> -		goto ERRORHANDLER;
> +		return -EFAULT;
>  	}
>  
>  	/* construct a new message */
>  	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
>  	if (!pstrMessage)
>  		return -ENOMEM;
> +
>  	pstrMessage->u32Length = u32SendBufferSize;
>  	pstrMessage->pstrNext = NULL;
>  	pstrMessage->pvBuffer = kmalloc(u32SendBufferSize, GFP_ATOMIC);
>  	if (!pstrMessage->pvBuffer) {
> -		result = -ENOMEM;
> -		goto ERRORHANDLER;
> +		kfree(pstrMessage);
> +		return -ENOMEM;
>  	}
>  	memcpy(pstrMessage->pvBuffer, pvSendBuffer, u32SendBufferSize);
>  
> @@ -102,15 +100,7 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>  	spin_unlock_irqrestore(&pHandle->strCriticalSection, flags);
>  
>  	up(&pHandle->hSem);
> -
> -ERRORHANDLER:
> -	/* error occured, free any allocations */
> -	if (pstrMessage) {
> -		kfree(pstrMessage->pvBuffer);
> -		kfree(pstrMessage);
> -	}
> -
> -	return result;
> +	return 0;

Aren't you now leaking memory as you aren't freeing pstrMessage and the
buffer on the "normal" return path?

thanks,

greg k-h

Re: [PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Dan Carpenter]

On Sun, Oct 04, 2015 at 09:44:57AM +0100, Greg KH wrote:
> Aren't you now leaking memory as you aren't freeing pstrMessage and the
> buffer on the "normal" return path?

It's supposed to.  It's a bug fix.  I explained to him in the first
version that his changelog sucks.

regards,
dan carpenter

Re: [PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[Chandra Gorentla]

On Sun, Oct 04, 2015 at 09:43:35AM +0100, Greg KH wrote:
> On Sat, Oct 03, 2015 at 02:57:29PM +0530, Chandra S Gorentla wrote:
> > The spin_lock_irqsave is moved to just beginning of critical section.
> > This change moves a couple of return statements out of the lock.
> > 
> > Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
> > ---
> >  drivers/staging/wilc1000/wilc_msgqueue.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
> > index d5ebd6d..284a3f5 100644
> > --- a/drivers/staging/wilc1000/wilc_msgqueue.c
> > +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
> > @@ -72,8 +72,6 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
> >  		goto ERRORHANDLER;
> >  	}
> >  
> > -	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
> > -
> >  	/* construct a new message */
> >  	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
> 
> As you have moved the lock, can you also change this to GFP_KERNEL as
> well because we do not have a lock held?
Can 'the change to GFP_KERNEL' be done in a separate patch?
The lock is to protect linked list manipulations; in this function items
are added to the list.
> 
> And how have you tested that this is ok?  What is this lock trying to
> protect?
I load this module on a notebook computer.   I added some code to
wilc_debugfs.c to invoke the functions in the file wilc_msgqueue.c
> 
> thanks,
> 
> greg k-h

Re: [PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Greg KH]

On Sun, Oct 04, 2015 at 12:16:31PM +0300, Dan Carpenter wrote:
> On Sun, Oct 04, 2015 at 09:44:57AM +0100, Greg KH wrote:
> > Aren't you now leaking memory as you aren't freeing pstrMessage and the
> > buffer on the "normal" return path?
> 
> It's supposed to.  It's a bug fix.  I explained to him in the first
> version that his changelog sucks.

Ok, well, it still sucks :)

And it's out of my queue, so I'll wait for the next spin of this
series...

greg k-h

Re: [PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[Greg KH]

On Sun, Oct 04, 2015 at 03:37:13PM +0530, Chandra Gorentla wrote:
> On Sun, Oct 04, 2015 at 09:43:35AM +0100, Greg KH wrote:
> > On Sat, Oct 03, 2015 at 02:57:29PM +0530, Chandra S Gorentla wrote:
> > > The spin_lock_irqsave is moved to just beginning of critical section.
> > > This change moves a couple of return statements out of the lock.
> > > 
> > > Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
> > > ---
> > >  drivers/staging/wilc1000/wilc_msgqueue.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
> > > index d5ebd6d..284a3f5 100644
> > > --- a/drivers/staging/wilc1000/wilc_msgqueue.c
> > > +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
> > > @@ -72,8 +72,6 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
> > >  		goto ERRORHANDLER;
> > >  	}
> > >  
> > > -	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
> > > -
> > >  	/* construct a new message */
> > >  	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
> > 
> > As you have moved the lock, can you also change this to GFP_KERNEL as
> > well because we do not have a lock held?
> Can 'the change to GFP_KERNEL' be done in a separate patch?

Yes.

> The lock is to protect linked list manipulations; in this function items
> are added to the list.

Ok, please add that description to the patch so we know what is going
on, and that you know what is going on as well :)

thanks,

greg k-h

Re: [PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Chandra Gorentla]

On Sun, Oct 04, 2015 at 09:44:57AM +0100, Greg KH wrote:
> On Sat, Oct 03, 2015 at 02:57:30PM +0530, Chandra S Gorentla wrote:
> >  - kfree is being called for the members of the queue without
> >    de-queuing them; they are just inserted within this function;
> >    they are supposed to be de-queued and freed in a function
> >    for receiving the queue items
> >  - goto statements are removed
> >  - After kfree correction, there is no need for target block
> >    of goto statement; hence it is removed
> > 
> > Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
> > ---
> >  drivers/staging/wilc1000/wilc_msgqueue.c | 22 ++++++----------------
> >  1 file changed, 6 insertions(+), 16 deletions(-)
> > 
> > diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
> > index 284a3f5..eae90be 100644
> > --- a/drivers/staging/wilc1000/wilc_msgqueue.c
> > +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
> > @@ -56,32 +56,30 @@ int wilc_mq_destroy(WILC_MsgQueueHandle *pHandle)
> >  int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
> >  			     const void *pvSendBuffer, u32 u32SendBufferSize)
> >  {
> > -	int result = 0;
> >  	unsigned long flags;
> >  	Message *pstrMessage = NULL;
> >  
> >  	if ((!pHandle) || (u32SendBufferSize == 0) || (!pvSendBuffer)) {
> >  		PRINT_ER("pHandle or pvSendBuffer is null\n");
> > -		result = -EFAULT;
> > -		goto ERRORHANDLER;
> > +		return -EFAULT;
> >  	}
> >  
> >  	if (pHandle->bExiting) {
> >  		PRINT_ER("pHandle fail\n");
> > -		result = -EFAULT;
> > -		goto ERRORHANDLER;
> > +		return -EFAULT;
> >  	}
> >  
> >  	/* construct a new message */
> >  	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
> >  	if (!pstrMessage)
> >  		return -ENOMEM;
> > +
> >  	pstrMessage->u32Length = u32SendBufferSize;
> >  	pstrMessage->pstrNext = NULL;
> >  	pstrMessage->pvBuffer = kmalloc(u32SendBufferSize, GFP_ATOMIC);
> >  	if (!pstrMessage->pvBuffer) {
> > -		result = -ENOMEM;
> > -		goto ERRORHANDLER;
> > +		kfree(pstrMessage);
> > +		return -ENOMEM;
> >  	}
> >  	memcpy(pstrMessage->pvBuffer, pvSendBuffer, u32SendBufferSize);
> >  
> > @@ -102,15 +100,7 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
> >  	spin_unlock_irqrestore(&pHandle->strCriticalSection, flags);
> >  
> >  	up(&pHandle->hSem);
> > -
> > -ERRORHANDLER:
> > -	/* error occured, free any allocations */
> > -	if (pstrMessage) {
> > -		kfree(pstrMessage->pvBuffer);
> > -		kfree(pstrMessage);
> > -	}
> > -
> > -	return result;
> > +	return 0;
> 
> Aren't you now leaking memory as you aren't freeing pstrMessage and the
> buffer on the "normal" return path?
In the normal path kfree is called in a separate (wilc_mq_recv) function.
The purpose of the currently modified function (wilc_mq_send) is to post
a message to a queue by allocating memory for the message.  The receiver
function is supposed to remove the message from the queue and free the
memory.

> 
> thanks,
> 
> greg k-h

Re: [PATCH 2/2] drivers: staging: wilc1000: Call kfree only for error cases

[Tony Cho]

On 2015년 10월 04일 19:28, Chandra Gorentla wrote:
> On Sun, Oct 04, 2015 at 09:44:57AM +0100, Greg KH wrote:
>> On Sat, Oct 03, 2015 at 02:57:30PM +0530, Chandra S Gorentla wrote:
>>>   - kfree is being called for the members of the queue without
>>>     de-queuing them; they are just inserted within this function;
>>>     they are supposed to be de-queued and freed in a function
>>>     for receiving the queue items
>>>   - goto statements are removed
>>>   - After kfree correction, there is no need for target block
>>>     of goto statement; hence it is removed
>>>
>>> Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
>>> ---
>>>   drivers/staging/wilc1000/wilc_msgqueue.c | 22 ++++++----------------
>>>   1 file changed, 6 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
>>> index 284a3f5..eae90be 100644
>>> --- a/drivers/staging/wilc1000/wilc_msgqueue.c
>>> +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
>>> @@ -56,32 +56,30 @@ int wilc_mq_destroy(WILC_MsgQueueHandle *pHandle)
>>>   int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>>>   			     const void *pvSendBuffer, u32 u32SendBufferSize)
>>>   {
>>> -	int result = 0;
>>>   	unsigned long flags;
>>>   	Message *pstrMessage = NULL;
>>>   
>>>   	if ((!pHandle) || (u32SendBufferSize == 0) || (!pvSendBuffer)) {
>>>   		PRINT_ER("pHandle or pvSendBuffer is null\n");
>>> -		result = -EFAULT;
>>> -		goto ERRORHANDLER;
>>> +		return -EFAULT;
>>>   	}
>>>   
>>>   	if (pHandle->bExiting) {
>>>   		PRINT_ER("pHandle fail\n");
>>> -		result = -EFAULT;
>>> -		goto ERRORHANDLER;
>>> +		return -EFAULT;
>>>   	}
>>>   
>>>   	/* construct a new message */
>>>   	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
>>>   	if (!pstrMessage)
>>>   		return -ENOMEM;
>>> +
>>>   	pstrMessage->u32Length = u32SendBufferSize;
>>>   	pstrMessage->pstrNext = NULL;
>>>   	pstrMessage->pvBuffer = kmalloc(u32SendBufferSize, GFP_ATOMIC);
>>>   	if (!pstrMessage->pvBuffer) {
>>> -		result = -ENOMEM;
>>> -		goto ERRORHANDLER;
>>> +		kfree(pstrMessage);
>>> +		return -ENOMEM;
>>>   	}
>>>   	memcpy(pstrMessage->pvBuffer, pvSendBuffer, u32SendBufferSize);
>>>   
>>> @@ -102,15 +100,7 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>>>   	spin_unlock_irqrestore(&pHandle->strCriticalSection, flags);
>>>   
>>>   	up(&pHandle->hSem);
>>> -
>>> -ERRORHANDLER:
>>> -	/* error occured, free any allocations */
>>> -	if (pstrMessage) {
>>> -		kfree(pstrMessage->pvBuffer);
>>> -		kfree(pstrMessage);
>>> -	}
>>> -
>>> -	return result;
>>> +	return 0;
>> Aren't you now leaking memory as you aren't freeing pstrMessage and the
>> buffer on the "normal" return path?
> In the normal path kfree is called in a separate (wilc_mq_recv) function.
> The purpose of the currently modified function (wilc_mq_send) is to post
> a message to a queue by allocating memory for the message.  The receiver
> function is supposed to remove the message from the queue and free the
> memory.
>
This patch is reasonable and normal free is done in recv function as Chandra said.

Thanks,

Tony.

>> thanks,
>>
>> greg k-h
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 1/2] drivers: staging: wilc1000: Move spin lock to the start of critical section

[Tony Cho]

On 2015년 10월 04일 17:43, Greg KH wrote:
> On Sat, Oct 03, 2015 at 02:57:29PM +0530, Chandra S Gorentla wrote:
>> The spin_lock_irqsave is moved to just beginning of critical section.
>> This change moves a couple of return statements out of the lock.
>>
>> Signed-off-by: Chandra S Gorentla <csgorentla@gmail.com>
>> ---
>>   drivers/staging/wilc1000/wilc_msgqueue.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/staging/wilc1000/wilc_msgqueue.c b/drivers/staging/wilc1000/wilc_msgqueue.c
>> index d5ebd6d..284a3f5 100644
>> --- a/drivers/staging/wilc1000/wilc_msgqueue.c
>> +++ b/drivers/staging/wilc1000/wilc_msgqueue.c
>> @@ -72,8 +72,6 @@ int wilc_mq_send(WILC_MsgQueueHandle *pHandle,
>>   		goto ERRORHANDLER;
>>   	}
>>   
>> -	spin_lock_irqsave(&pHandle->strCriticalSection, flags);
>> -
>>   	/* construct a new message */
>>   	pstrMessage = kmalloc(sizeof(Message), GFP_ATOMIC);
> As you have moved the lock, can you also change this to GFP_KERNEL as
> well because we do not have a lock held?
>
> And how have you tested that this is ok?  What is this lock trying to
> protect?

This function is called even in interrupt context, so GFP_ATOMIC should be called. The spinlock
also should protect pstrMessage from allocating the memory, so we don't place it to the beginning
of critical section as Chandra said.

Thanks,
Tony.

> thanks,
>
> greg k-h
> --
> To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html