Re: [PATCH/RFC] SPI: add async message handing library to David Brownell's core

[Stephen Street <stephen@streetfiresound.com>]

On Tue, 2005-12-13 at 17:06 +0300, Vitaly Wool wrote:
> Greetings,
> 
> This thingie hasn't been thoroughly tested yet, but it's lightweight
> and easy to understand so I don't think solving the problems that 
> may arise will take long. Though I haven't actually done that yet, 
> I'm sure that Stephen's PXA SSP driver will become easier to understand 
> and less in footprint and will work faster when it's rewritten using 
> this library. (Yes, I do expect performance improvement here as the 
> current implementation schedules multiple tasklets, so 
> scheduling penalty is high.)

Is this really true?  Is tasklet scheduling "harder" than kernal thread
scheduling?  A close look at my PXA SSP SPI implementation will reveal
that my design is nearly lock-less and callable from any execution
context (i.e. interrupt context).

In general, the "pump_transfer" and "pump_message" tasklets should
mutually exclusive (anything else is a programming
mistake). I certainly could collapse the tasklets into a single one if
you think this would be better?

> + * spi_queue - (internal) queue the message to be processed asynchronously
> + * @spi: SPI device to perform transfer to/from
> + * @msg: message to be sent
> + * Description:
> + * 	This function queues the message to SPI controller's queue.
> + */
> +static int spi_queue(struct spi_device *spi, struct spi_message *msg)
> +{
> +	struct threaded_async_data *td = spi->master->context;
> +	int rc = 0;
> +
> +	if (!td) {
> +		rc = -EINVAL;
> +		goto out;
> +	}
> +
> +	msg->spi = spi;
> +	down(&td->lock);
> +	list_add_tail(&msg->queue, &td->msgs);
> +	dev_dbg(spi->dev.parent, "message has been queued\n");
> +	up(&td->lock);
> +	wake_up_interruptible(&td->wq);
> +
> +out:
> +	return rc;
> +}
> +

This can not be invoke this from "interrupt context" which is a
requirement for my SPI devices (CS8415A, CS8405A, CS4341).