OK, I made some time to look at that.  As I've mentioned already,
the "spi_bitbang" code needed to morph in this direction, and
that was in fact the patch I was hoping you'd be sending.  So I
finished that up instead, and have sent it separately ... but
I've also attached the C file to this response.


> +struct threaded_async_data {
> +	atomic_t exiting;
> +	struct spi_master *master;
> +	struct task_struct *thread;
> +	wait_queue_head_t wq;

I just kept the workqueue, named after its device.
It makes the code look a lot simpler!


> +	struct list_head msgs;
> +	spinlock_t lock;

> +	int (*xfer) (struct spi_master *, struct spi_message *);
> +};
> +
> +/**
> + * spi_start_async - start the thread
> + * @master: SPI controller structure which the thread is related to
> + * @return: abstract pointer to the thread context
> + */
> +int spi_start_async (struct spi_master *master, int (*xfer)(struct spi_master *,
> 				struct spi_message *)) 

I did this differently, but liked your start/stop names.  :)
So:
	int spi_bitbang_start(struct spi_bitbang *bitbang);
	int spi_bitbang_stop(struct spi_bitbang *bitbang);

to start and stop processing the queue associated with the bitbanged
spi_master.  Callbacks just get stored in the structure; simpler,
and the return value is just a normal zero-or-negative-errno.


> @@ -152,6 +152,7 @@ static inline void spi_unregister_driver
>   * 	device's SPI controller; protocol code may call this.
>   * @transfer: adds a message to the controller's transfer queue.
>   * @cleanup: frees controller-specific state
> + * @context: controller-specific data
>   *
>   * Each SPI master controller can communicate with one or more spi_device
>   * children.  These make a small bus, sharing MOSI, MISO and SCK signals

I've not seen a need for that yet; the class_get_devdata() is doing
that already.  And it's now wrapped up as spi_master_get_devdata(). 

- Dave