Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1445547

Re: [PATCH v8 2/4] i2c-smbus: add SMBus Host Notify support

Show all headers | View raw

Hi Benjamin, Wolfram,

Now that I have reviewed the i2c-i801 part of the implementation, I'm
wondering...

On Thu,  9 Jun 2016 16:53:48 +0200, Benjamin Tissoires wrote:
> +/**
> + * i2c_setup_smbus_host_notify - Allocate a new smbus_host_notify for the given
> + * I2C adapter.
> + * @adapter: the adapter we want to associate a Host Notify function
> + *
> + * Returns a struct smbus_host_notify pointer on success, and NULL on failure.
> + * The resulting smbus_host_notify must not be freed afterwards, it is a
> + * managed resource already.
> + */
> +struct smbus_host_notify *i2c_setup_smbus_host_notify(struct i2c_adapter *adap)
> +{
> +	struct smbus_host_notify *host_notify;
> +
> +	host_notify = devm_kzalloc(&adap->dev, sizeof(struct smbus_host_notify),
> +				   GFP_KERNEL);
> +	if (!host_notify)
> +		return NULL;
> +
> +	host_notify->adapter = adap;
> +
> +	spin_lock_init(&host_notify->lock);
> +	INIT_WORK(&host_notify->work, smbus_host_notify_work);

Here we initialize a workqueue.

> +
> +	return host_notify;
> +}
> +EXPORT_SYMBOL_GPL(i2c_setup_smbus_host_notify);
> +
> +/**
> + * i2c_handle_smbus_host_notify - Forward a Host Notify event to the correct
> + * I2C client.
> + * @host_notify: the struct host_notify attached to the relevant adapter
> + * @data: the Host Notify data which contains the payload and address of the
> + * client
> + * Context: can't sleep
> + *
> + * Helper function to be called from an I2C bus driver's interrupt
> + * handler. It will schedule the Host Notify work, in turn calling the
> + * corresponding I2C device driver's alert function.
> + *
> + * host_notify should be a valid pointer previously returned by
> + * i2c_setup_smbus_host_notify().
> + */
> +int i2c_handle_smbus_host_notify(struct smbus_host_notify *host_notify,
> +				 unsigned short addr, unsigned int data)
> +{
> +	unsigned long flags;
> +	struct i2c_adapter *adapter;
> +
> +	if (!host_notify || !host_notify->adapter)
> +		return -EINVAL;
> +
> +	adapter = host_notify->adapter;
> +
> +	spin_lock_irqsave(&host_notify->lock, flags);
> +
> +	if (host_notify->pending) {
> +		spin_unlock_irqrestore(&host_notify->lock, flags);
> +		dev_warn(&adapter->dev, "Host Notify already scheduled.\n");
> +		return -EBUSY;
> +	}
> +
> +	host_notify->payload = data;
> +	host_notify->addr = addr;
> +
> +	/* Mark that there is a pending notification and release the lock */
> +	host_notify->pending = true;
> +	spin_unlock_irqrestore(&host_notify->lock, flags);
> +
> +	return schedule_work(&host_notify->work);

And here we use it.

> +}
> +EXPORT_SYMBOL_GPL(i2c_handle_smbus_host_notify);

But what happens on i2c_adapter removal? What prevents the following
sequence from happening?

1* A Host Notify event happens.
2* The event is handled and queued by i2c_handle_smbus_host_notify().
3* Someone tears down the underlying i2c_adapter (for example "rmmod
   i2c-i801".)
4* The workqueue is processed, accessing memory which has already been
   freed.

Of course it would be back luck, but that's pretty much the definition
of a race condition ;-)

To be on the safe side, don't we need a teardown function in i2c-smbus,
that could be called before i2c_del_adapter, which would remove the
host notify handle and flush the workqueue?

-- 
Jean Delvare
SUSE L3 Support

Back to linux.kernel | Previous | Next — Next in thread | Find similar | Unroll thread

Thread

Re: [PATCH v8 2/4] i2c-smbus: add SMBus Host Notify support Jean Delvare <jdelvare@suse.de> - 2016-07-18 16:40 +0200
  Re: [PATCH v8 2/4] i2c-smbus: add SMBus Host Notify support Benjamin Tissoires <benjamin.tissoires@redhat.com> - 2016-07-18 18:40 +0200
    Re: [PATCH v8 2/4] i2c-smbus: add SMBus Host Notify support Jean Delvare <jdelvare@suse.de> - 2016-07-18 22:50 +0200

csiph-web