Re: [PATCH] char_dev: replace cdev_map with an xarray

[Matthew Wilcox <willy@infradead.org>]

On Mon, Jan 11, 2021 at 06:05:13PM +0100, Christoph Hellwig wrote:
> None of the complicated overlapping regions bits of the kobj_map are
> required for the character device lookup, so just a trivial xarray
> instead.

Thanks for doing this.  We could make it more efficient for chardevs
that occupy 64 or more consecutive/aligned devices -- is it worth doing?

> +static struct cdev *cdev_lookup(dev_t dev)
> +{
> +	struct cdev *cdev;
> +
> +	mutex_lock(&chrdevs_lock);
> +	cdev = xa_load(&cdev_map, dev);
> +	if (!cdev) {
> +		mutex_unlock(&chrdevs_lock);
> +		if (request_module("char-major-%d-%d",
> +				   MAJOR(dev), MINOR(dev)) > 0)
> +			/* Make old-style 2.4 aliases work */
> +			request_module("char-major-%d", MAJOR(dev));
> +		mutex_lock(&chrdevs_lock);
> +
> +		cdev = xa_load(&cdev_map, dev);
> +	}
> +	if (cdev && !cdev_get(cdev))
> +		cdev = NULL;
> +	mutex_unlock(&chrdevs_lock);
> +	return cdev;

What does the mutex protect here?  Is it cdev being freed?

> @@ -593,11 +601,16 @@ static void cdev_unmap(dev_t dev, unsigned count)
>   */
>  void cdev_del(struct cdev *p)
>  {
> -	cdev_unmap(p->dev, p->count);
> +	int i;
> +
> +	mutex_lock(&chrdevs_lock);
> +	for (i = 0; i < p->count; i++)
> +		xa_erase(&cdev_map, p->dev + i);
> +	mutex_unlock(&chrdevs_lock);

I don't understand what it's protecting here.  It's clearly not cdev_get
as that could happen before we acquire the mutex.  This also suggests
I should add an xa_erase_range() to the API.

But there's nothing wrong here, just some places that maybe could be
better, so:

Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>