[PATCH] mfd: viperboard: allocate I/O buffer separately

[PATCH] mfd: viperboard: allocate I/O buffer separately

[Octavian Purdila]

Currently the I/O buffer is allocated part of the device status
structure, potentially sharing the same cache line with other members
in this structure.

Allocate the buffer separately, to avoid the I/O operations corrupting
the device status structure due to cache line sharing.

Compiled tested only, as I don't have access to hardware.

Signed-off-by: Octavian Purdila <octavian.purdila@intel.com>
---
 drivers/mfd/viperboard.c       | 16 ++++++++++------
 include/linux/mfd/viperboard.h |  2 +-
 2 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/drivers/mfd/viperboard.c b/drivers/mfd/viperboard.c
index e00f534..d27c131 100644
--- a/drivers/mfd/viperboard.c
+++ b/drivers/mfd/viperboard.c
@@ -59,9 +59,13 @@ static int vprbrd_probe(struct usb_interface *interface,
 
 	/* allocate memory for our device state and initialize it */
 	vb = kzalloc(sizeof(*vb), GFP_KERNEL);
-	if (vb == NULL) {
-		dev_err(&interface->dev, "Out of memory\n");
+	if (vb == NULL)
 		return -ENOMEM;
+
+	vb->buf = kzalloc(sizeof(struct vprbrd_i2c_write_msg), GFP_KERNEL);
+	if (vb->buf == NULL) {
+		ret = -ENOMEM;
+		goto error;
 	}
 
 	mutex_init(&vb->lock);
@@ -103,10 +107,9 @@ static int vprbrd_probe(struct usb_interface *interface,
 	return 0;
 
 error:
-	if (vb) {
-		usb_put_dev(vb->usb_dev);
-		kfree(vb);
-	}
+	usb_put_dev(vb->usb_dev);
+	kfree(vb->buf);
+	kfree(vb);
 
 	return ret;
 }
@@ -118,6 +121,7 @@ static void vprbrd_disconnect(struct usb_interface *interface)
 	mfd_remove_devices(&interface->dev);
 	usb_set_intfdata(interface, NULL);
 	usb_put_dev(vb->usb_dev);
+	kfree(vb->buf);
 	kfree(vb);
 
 	dev_dbg(&interface->dev, "disconnected\n");
diff --git a/include/linux/mfd/viperboard.h b/include/linux/mfd/viperboard.h
index 1934528..af928d0 100644
--- a/include/linux/mfd/viperboard.h
+++ b/include/linux/mfd/viperboard.h
@@ -103,7 +103,7 @@ struct vprbrd_i2c_addr_msg {
 struct vprbrd {
 	struct usb_device *usb_dev; /* the usb device for this device */
 	struct mutex lock;
-	u8 buf[sizeof(struct vprbrd_i2c_write_msg)];
+	u8 *buf;
 	struct platform_device pdev;
 };
 
-- 
1.9.1

Re: [PATCH] mfd: viperboard: allocate I/O buffer separately

[Johan Hovold]

On Mon, Sep 22, 2014 at 05:46:52PM +0300, Octavian Purdila wrote:
> Currently the I/O buffer is allocated part of the device status
> structure, potentially sharing the same cache line with other members
> in this structure.
> 
> Allocate the buffer separately, to avoid the I/O operations corrupting
> the device status structure due to cache line sharing.
> 
> Compiled tested only, as I don't have access to hardware.
> 
> Signed-off-by: Octavian Purdila <octavian.purdila@intel.com>
> ---

Change itself looks sane, although the driver's use of a shared buffer
and relying on undocumented locking is a different story.

However, you do more than your commit message claims below.

>  drivers/mfd/viperboard.c       | 16 ++++++++++------
>  include/linux/mfd/viperboard.h |  2 +-
>  2 files changed, 11 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/mfd/viperboard.c b/drivers/mfd/viperboard.c
> index e00f534..d27c131 100644
> --- a/drivers/mfd/viperboard.c
> +++ b/drivers/mfd/viperboard.c
> @@ -59,9 +59,13 @@ static int vprbrd_probe(struct usb_interface *interface,
>  
>  	/* allocate memory for our device state and initialize it */
>  	vb = kzalloc(sizeof(*vb), GFP_KERNEL);
> -	if (vb == NULL) {
> -		dev_err(&interface->dev, "Out of memory\n");

Here you're also removing a redundant OOM message.

> +	if (vb == NULL)
>  		return -ENOMEM;
> +
> +	vb->buf = kzalloc(sizeof(struct vprbrd_i2c_write_msg), GFP_KERNEL);
> +	if (vb->buf == NULL) {
> +		ret = -ENOMEM;
> +		goto error;
>  	}
>  
>  	mutex_init(&vb->lock);
> @@ -103,10 +107,9 @@ static int vprbrd_probe(struct usb_interface *interface,
>  	return 0;
>  
>  error:
> -	if (vb) {

And cleaning up the error path.

> -		usb_put_dev(vb->usb_dev);
> -		kfree(vb);
> -	}
> +	usb_put_dev(vb->usb_dev);
> +	kfree(vb->buf);
> +	kfree(vb);
>  
>  	return ret;
>  }

Don't mix fixes and clean ups like this, but rather submit them as
separate patches.

Johan

Re: [PATCH] mfd: viperboard: allocate I/O buffer separately

[Octavian Purdila]

On Mon, Sep 22, 2014 at 7:08 PM, Johan Hovold <johan@kernel.org> wrote:
> On Mon, Sep 22, 2014 at 05:46:52PM +0300, Octavian Purdila wrote:
>> Currently the I/O buffer is allocated part of the device status
>> structure, potentially sharing the same cache line with other members
>> in this structure.
>>
>> Allocate the buffer separately, to avoid the I/O operations corrupting
>> the device status structure due to cache line sharing.
>>
>> Compiled tested only, as I don't have access to hardware.
>>
>> Signed-off-by: Octavian Purdila <octavian.purdila@intel.com>
>> ---
>
> Change itself looks sane, although the driver's use of a shared buffer
> and relying on undocumented locking is a different story.
>
> However, you do more than your commit message claims below.
>
>>  drivers/mfd/viperboard.c       | 16 ++++++++++------
>>  include/linux/mfd/viperboard.h |  2 +-
>>  2 files changed, 11 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/mfd/viperboard.c b/drivers/mfd/viperboard.c
>> index e00f534..d27c131 100644
>> --- a/drivers/mfd/viperboard.c
>> +++ b/drivers/mfd/viperboard.c
>> @@ -59,9 +59,13 @@ static int vprbrd_probe(struct usb_interface *interface,
>>
>>       /* allocate memory for our device state and initialize it */
>>       vb = kzalloc(sizeof(*vb), GFP_KERNEL);
>> -     if (vb == NULL) {
>> -             dev_err(&interface->dev, "Out of memory\n");
>
> Here you're also removing a redundant OOM message.
>
>> +     if (vb == NULL)
>>               return -ENOMEM;
>> +
>> +     vb->buf = kzalloc(sizeof(struct vprbrd_i2c_write_msg), GFP_KERNEL);
>> +     if (vb->buf == NULL) {
>> +             ret = -ENOMEM;
>> +             goto error;
>>       }
>>
>>       mutex_init(&vb->lock);
>> @@ -103,10 +107,9 @@ static int vprbrd_probe(struct usb_interface *interface,
>>       return 0;
>>
>>  error:
>> -     if (vb) {
>
> And cleaning up the error path.
>
>> -             usb_put_dev(vb->usb_dev);
>> -             kfree(vb);
>> -     }
>> +     usb_put_dev(vb->usb_dev);
>> +     kfree(vb->buf);
>> +     kfree(vb);
>>
>>       return ret;
>>  }
>
> Don't mix fixes and clean ups like this, but rather submit them as
> separate patches.
>

Fair enough. Is it OK to send all of the cleanups in a single separate patch?

Re: [PATCH] mfd: viperboard: allocate I/O buffer separately

[Johan Hovold]

On Mon, Sep 22, 2014 at 07:19:37PM +0300, Octavian Purdila wrote:
> On Mon, Sep 22, 2014 at 7:08 PM, Johan Hovold <johan@kernel.org> wrote:
> > On Mon, Sep 22, 2014 at 05:46:52PM +0300, Octavian Purdila wrote:

> > Don't mix fixes and clean ups like this, but rather submit them as
> > separate patches.
> >
> Fair enough. Is it OK to send all of the cleanups in a single separate
> patch?

In this case, I'd say so. But do the clean-ups on top of the minimal fix
to facilitate back-porting if someone ever decides that that is needed.

Johan