Re: [PATCH 2/9] x86/amd-iommu: Introduce global dev_data_list

[Chris Wright <chrisw@sous-sol.org>]

* Joerg Roedel (joerg.roedel@amd.com) wrote:
> +static struct iommu_dev_data *alloc_dev_data(void)
> +{
> +	struct iommu_dev_data *dev_data;
> +	unsigned long flags;
> +
> +	dev_data = kzalloc(sizeof(*dev_data), GFP_KERNEL);
> +	if (!dev_data)
> +		return NULL;
> +
> +	atomic_set(&dev_data->bind, 0);
> +
> +	spin_lock_irqsave(&dev_data_list_lock, flags);
> +	list_add_tail(&dev_data->dev_data_list, &dev_data_list);
> +	spin_unlock_irqrestore(&dev_data_list_lock, flags);

Globally visible but only paritially initiailized.  I didn't see any, but
would this ever cause an issue?

> +static void free_dev_data(struct iommu_dev_data *dev_data)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&dev_data_list_lock, flags);
> +	list_del(&dev_data->dev_data_list);
> +	spin_unlock_irqrestore(&dev_data_list_lock, flags);
> +
> +	kfree(dev_data);
<snip>
> +	/* Free all of our dev_data structures */
> +	list_for_each_entry_safe(dev_data, n, &dev_data_list, dev_data_list)
> +		free_dev_data(dev_data);

Given that it's not actually contended in early init, should be fine...but
typically full list traversal would be protected by lock rather than
repeatedly acquiring and releasing the lock.