[PATCH 023/141] staging: unisys: add visorbus driver

[Dan Carpenter]

On Tue, May 05, 2015 at 06:36:00PM -0400, Benjamin Romer wrote:
> +/** This is called automatically upon adding a visor_device (device_add), or
> + *  adding a visor_driver (visorbus_register_visor_driver), but only after
> + *  visorbus_match has returned 1 to indicate a successful match between
> + *  driver and device.
> + */
> +static int
> +visordriver_probe_device(struct device *xdev)
> +{
> +	int rc;
> +	struct visor_driver *drv;
> +	struct visor_device *dev;
> +
> +	drv = to_visor_driver(xdev->driver);
> +	dev = to_visor_device(xdev);
> +	down(&dev->visordriver_callback_lock);
> +	dev->being_removed = FALSE;
> +	/*
> +	 * ensure that the dev->being_removed flag is cleared before
> +	 * we start the probe
> +	 */
> +	wmb();
> +	get_device(&dev->device);
> +	if (!drv->probe) {
> +		up(&dev->visordriver_callback_lock);
> +		rc = -1;
> +		goto away;
> +	}
> +	rc = drv->probe(dev);
> +	if (rc < 0)
> +		goto away;
> +
> +	fix_vbus_dev_info(dev);
> +	up(&dev->visordriver_callback_lock);
> +	rc = 0;
> +away:
> +	if (rc != 0)
> +		put_device(&dev->device);
> +	/*  We could get here more than once if the child driver module is
> +	 *  unloaded and re-loaded while devices are present.  That's why we
> +	 *  need a flag to be sure that we only respond to the device_create
> +	 *  once.  We cannot respond to the device_create prior to here,
> +	 *  because until we call drv->probe() above, the channel has not been
> +	 *  initialized.
> +	 */
> +	if (!dev->responded_to_device_create) {
> +		dev->responded_to_device_create = TRUE;
> +		if (chipset_responders.device_create)
> +			(*chipset_responders.device_create)(dev->chipset_bus_no,
> +							    dev->chipset_dev_no,
> +							    rc);
> +	}
> +	return rc;
> +}

The error handling in this function looks kind of weird.  Also if
drv->probe() fails then we need to unlock.  One err bug.

regards,
dan carpenter