Documentation/spi/spi-summary

Overview of Linux kernel SPI support
====================================

02-Dec-2005

What is SPI?
------------
The "Serial Peripheral Interface" (SPI) is a synchronous four wire serial
link used to connect microcontrollers to sensors, memory, and peripherals.

The three signal wires hold a clock (SCLK, often on the order of 10 MHz),
and parallel data lines with "Master Out, Slave In" (MOSI) or "Master In,
Slave Out" (MISO) signals.  (Other names are also used.)  There are four
clocking modes through which data is exchanged; mode-0 and mode-3 are most
commonly used.  Each clock cycle shifts data out and data in; the clock
doesn't cycle except when there is data to shift.

SPI masters may use a "chip select" line to activate a given SPI slave
device, so those three signal wires may be connected to several chips
in parallel.  All SPI slaves support chipselects.  Some devices have
other signals, often including an interrupt to the master.

Unlike serial busses like USB or SMBUS, even low level protocols for
SPI slave functions are usually not interoperable between vendors
(except for cases like SPI memory chips).

  - SPI may be used for request/response style device protocols, as with
    touchscreen sensors and memory chips.

  - It may also be used to stream data in either direction (half duplex),
    or both of them at the same time (full duplex).

  - Some devices may use eight bit words.  Others may different word
    lengths, such as streams of 12-bit or 20-bit digital samples.

In the same way, SPI slaves will only rarely support any kind of automatic
discovery/enumeration protocol.  The tree of slave devices accessible from
a given SPI master will normally be set up manually, with configuration
tables.

SPI is only one of the names used by such four-wire protocols, and
most controllers have no problem handling "MicroWire" (think of it as
half-duplex SPI, for request/response protocols), SSP ("Synchronous
Serial Protocol"), PSP ("Programmable Serial Protocol"), and other
related protocols.

Microcontrollers often support both master and slave sides of the SPI
protocol.  This document (and Linux) currently only supports the master
side of SPI interactions.


Who uses it?  On what kinds of systems?
---------------------------------------
Linux developers using SPI are probably writing device drivers for embedded
systems boards.  SPI is used to control external chips, and it is also a
protocol supported by every MMC or SD memory card.  (The older "DataFlash"
cards, predating MMC cards but using the same connectors and card shape,
support only SPI.)  Some PC hardware uses SPI flash for BIOS code.

SPI slave chips range from digital/analog converters used for analog
sensors and codecs, to memory, to peripherals like USB controllers
or Ethernet adapters; and more.

Most systems using SPI will integrate a few devices on a mainboard.
Some provide SPI links on expansion connectors; in cases where no
dedicated SPI controller exists, GPIO pins can be used to create a
low speed "bitbanging" adapter.  Very few systems will "hotplug" an SPI
controller; the reasons to use SPI focus on low cost and simple operation,
and if dynamic reconfiguration is important, USB will often be a more
appropriate low-pincount peripheral bus.

Many microcontrollers that can run Linux integrate one or more I/O
interfaces with SPI modes.  Given SPI support, they could use MMC or SD
cards without needing a special purpose MMC/SD/SDIO controller.


How do these driver programming interfaces work?
------------------------------------------------
The <linux/spi/spi.h> header file includes kerneldoc, as does the
main source code, and you should certainly read that.  This is just
an overview, so you get the big picture before the details.

SPI requests always go into I/O queues.  Requests for a given SPI device
are always executed in FIFO order, and complete asynchronously through
completion callbacks.  There are also some simple synchronous wrappers
for those calls, including ones for common transaction types like writing
a command and then reading its response.

There are two types of SPI driver, here called:

  Controller drivers ... these are often built in to System-On-Chip
	processors, and often support both Master and Slave roles.
	These drivers touch hardware registers and may use DMA.
	Or they can be PIO bitbangers, needing just GPIO pins.

  Protocol drivers ... these pass messages through the controller
	driver to communicate with a Slave or Master device on the
	other side of an SPI link.

So for example one protocol driver might talk to the MTD layer to export
data to filesystems stored on SPI flash like DataFlash; and others might
control audio interfaces, present touchscreen sensors as input interfaces,
or monitor temperature and voltage levels during industrial processing.
And those might all be sharing the same controller driver.

A "struct spi_device" encapsulates the master-side interface between
those two types of driver.  At this writing, Linux has no slave side
programming interface.

There is a minimal core of SPI programming interfaces, focussing on
using driver model to connect controller and protocol drivers using
device tables provided by board specific initialization code.  SPI
shows up in sysfs in several locations:

   /sys/devices/.../CTLR/spiB.C ... spi_device for on bus "B",
	chipselect C, accessed through CTLR.

   /sys/bus/spi/devices/spiB.C ... symlink to the physical
   	spiB-C device

   /sys/bus/spi/drivers/D ... driver for one or more spi*.* devices

   /sys/class/spi_master/spiB ... class device for the controller
	managing bus "B".  All the spiB.* devices share the same
	physical SPI bus segment, with SCLK, MOSI, and MISO.


How does board-specific init code declare SPI devices?
------------------------------------------------------
Linux needs several kinds of information to properly configure SPI devices.
That information is normally provided by board-specific code, even for
chips that do support some of automated discovery/enumeration.

DECLARE CONTROLLERS

The first kind of information is a list of what SPI controllers exist.
For System-on-Chip (SOC) based boards, these will usually be platform
devices, and the controller may need some platform_data in order to
operate properly.  The "struct platform_device" will include resources
like the physical address of the controller's first register and its IRQ.

Platforms will often abstract the "register SPI controller" operation,
maybe coupling it with code to initialize pin configurations, so that
the arch/.../mach-*/board-*.c files for several boards can all share the
same basic controller setup code.  This is because most SOCs have several
SPI-capable controllers, and only the ones actually usable on a given
board should normally be set up and registered.

So for example arch/.../mach-*/board-*.c files might have code like:

	#include <asm/arch/spi.h>	/* for mysoc_spi_data */

	/* if your mach-* infrastructure doesn't support kernels that can
	 * run on multiple boards, pdata wouldn't benefit from "__init".
	 */
	static struct mysoc_spi_data __init pdata = { ... };

	static __init board_init(void)
	{
		...
		/* this board only uses SPI controller #2 */
		mysoc_register_spi(2, &pdata);
		...
	}

And SOC-specific utility code might look something like:

	#include <asm/arch/spi.h>

	static struct platform_device spi2 = { ... };

	void mysoc_register_spi(unsigned n, struct mysoc_spi_data *pdata)
	{
		struct mysoc_spi_data *pdata2;

		pdata2 = kmalloc(sizeof *pdata2, GFP_KERNEL);
		*pdata2 = pdata;
		...
		if (n == 2) {
			spi2->dev.platform_data = pdata2;
			register_platform_device(&spi2);

			/* also: set up pin modes so the spi2 signals are
			 * visible on the relevant pins ... bootloaders on
			 * production boards may already have done this, but
			 * developer boards will often need Linux to do it.
			 */
		}
		...
	}

Notice how the platform_data for boards may be different, even if the
same SOC controller is used.  For example, on one board SPI might use
an external clock, where another derives the SPI clock from current
settings of some master clock.


DECLARE SLAVE DEVICES

The second kind of information is a list of what SPI slave devices exist
on the target board, often with some board-specific data needed for the
driver to work correctly.

Normally your arch/.../mach-*/board-*.c files would provide a small table
listing the SPI devices on each board.  (This would typically be only a
small handful.)  That might look like:

	static struct ads7846_platform_data ads_info = {
		.vref_delay_usecs	= 100,
		.x_plate_ohms		= 580,
		.y_plate_ohms		= 410,
	};

	static struct spi_board_info spi_board_info[] __initdata = {
	{
		.modalias	= "ads7846",
		.platform_data	= &ads_info,
		.mode		= SPI_MODE_0,
		.irq		= GPIO_IRQ(31),
		.max_speed_hz	= 120000 /* max sample rate at 3V */ * 16,
		.bus_num	= 1,
		.chip_select	= 0,
	},
	};

Again, notice how board-specific information is provided; each chip may need
several types.  This example shows generic constraints like the fastest SPI
clock to allow (a function of board voltage in this case) or how an IRQ pin
is wired, plus chip-specific constraints like an important delay that's
changed by the capacitance at one pin.

(There's also "controller_data", information that may be useful to the
controller driver.  An example would be peripheral-specific DMA tuning
data or chipselect callbacks.  This is stored in spi_device later.)

The board_info should provide enough information to let the system work
without the chip's driver being loaded.  The most troublesome aspect of
that is likely the SPI_CS_HIGH bit in the spi_device.mode field, since
sharing a bus with a device that interprets chipselect "backwards" is
not possible.

Then your board initialization code would register that table with the SPI
infrastructure, so that it's available later when the SPI master controller
driver is registered:

	spi_register_board_info(spi_board_info, ARRAY_SIZE(spi_board_info));

Like with other static board-specific setup, you won't unregister those.


NON-STATIC CONFIGURATIONS

Developer boards often play by different rules than product boards, and one
example is the potential need to hotplug SPI devices and/or controllers.

For those cases you might need to use use spi_busnum_to_master() to look
up the spi bus master, and will likely need spi_new_device() to provide the
board info based on the board that was hotplugged.  Of course, you'd later
call at least spi_unregister_device() when that board is removed.


How do I write an "SPI Protocol Driver"?
----------------------------------------
All SPI drivers are currently kernel drivers.  A userspace driver API
would just be another kernel driver, probably offering some lowlevel
access through aio_read(), aio_write(), and ioctl() calls and using the
standard userspace sysfs mechanisms to bind to a given SPI device.

SPI protocol drivers somewhat resemble platform device drivers:

	static struct spi_driver CHIP_driver = {
		.driver = {
			.name		= "CHIP",
			.bus		= &spi_bus_type,
			.owner		= THIS_MODULE,
		},

		.probe		= CHIP_probe,
		.remove		= __devexit_p(CHIP_remove),
		.suspend	= CHIP_suspend,
		.resume		= CHIP_resume,
	};

The driver core will autmatically attempt to bind this driver to any SPI
device whose board_info gave a modalias of "CHIP".  Your probe() code
might look like this unless you're creating a class_device:

	static int __devinit CHIP_probe(struct spi_device *spi)
	{
		struct CHIP			*chip;
		struct CHIP_platform_data	*pdata;

		/* assuming the driver requires board-specific data: */
		pdata = &spi->dev.platform_data;
		if (!pdata)
			return -ENODEV;

		/* get memory for driver's per-chip state */
		chip = kzalloc(sizeof *chip, GFP_KERNEL);
		if (!chip)
			return -ENOMEM;
		dev_set_drvdata(&spi->dev, chip);

		... etc
		return 0;
	}

As soon as it enters probe(), the driver may issue I/O requests to
the SPI device using "struct spi_message".  When remove() returns,
the driver guarantees that it won't submit any more such messages.

  - An spi_message is a sequence of of protocol operations, executed
    as one atomic sequence.  SPI driver controls include:

      + when bidirectional reads and writes start ... by how its
        sequence of spi_transfer requests is arranged;

      + optionally defining short delays after transfers ... using
        the spi_transfer.delay_usecs setting;

      + whether the chipselect becomes inactive after a transfer and
        any delay ... by using the spi_transfer.cs_change flag;

      + hinting whether the next message is likely to go to this same
        device ... using the spi_transfer.cs_change flag on the last
	transfer in that atomic group, and potentially saving costs
	for chip deselect and select operations.

  - Follow standard kernel rules, and provide DMA-safe buffers in
    your messages.  That way controller drivers using DMA aren't forced
    to make extra copies unless the hardware requires it (e.g. working
    around hardware errata that force the use of bounce buffering).

    If standard dma_map_single() handling of these buffers is inappropriate,
    you can use spi_message.is_dma_mapped to tell the controller driver
    that you've already provided the relevant DMA addresses.

  - The basic I/O primitive is spi_async().  Async requests may be
    issued in any context (irq handler, task, etc) and completion
    is reported using a callback provided with the message.
    After any detected error, the chip is deselected and processing
    of that spi_message is aborted.

  - There are also synchronous wrappers like spi_sync(), and wrappers
    like spi_read(), spi_write(), and spi_write_then_read().  These
    may be issued only in contexts that may sleep, and they're all
    clean (and small, and "optional") layers over spi_async().

  - The spi_write_then_read() call, and convenience wrappers around
    it, should only be used with small amounts of data where the
    cost of an extra copy may be ignored.  It's designed to support
    common RPC-style requests, such as writing an eight bit command
    and reading a sixteen bit response -- spi_w8r16() being one its
    wrappers, doing exactly that.

Some drivers may need to modify spi_device characteristics like the
transfer mode, wordsize, or clock rate.  This is done with spi_setup(),
which would normally be called from probe() before the first I/O is
done to the device.

While "spi_device" would be the bottom boundary of the driver, the
upper boundaries might include sysfs (especially for sensor readings),
the input layer, ALSA, networking, MTD, the character device framework,
or other Linux subsystems.

Note that there are two types of memory your driver must manage as part
of interacting with SPI devices.

  - I/O buffers use the usual Linux rules, and must be DMA-safe.
    You'd normally allocate them from the heap or free page pool.
    Don't use the stack, or anything that's declared "static".

  - The spi_message and spi_transfer metadata used to glue those
    I/O buffers into a group of protocol transactions.  These can
    be allocated anywhere it's convenient, including as part of
    other allocate-once driver data structures.  Zero-init these.

If you like, spi_message_alloc() and spi_message_free() convenience
routines are available to allocate and zero-initialize an spi_message
with several transfers.


How do I write an "SPI Master Controller Driver"?
-------------------------------------------------
An SPI controller will probably be registered on the platform_bus; write
a driver to bind to the device, whichever bus is involved.

The main task of this type of driver is to provide an "spi_master".
Use spi_alloc_master() to allocate the master, and class_get_devdata()
to get the driver-private data allocated for that device.

	struct spi_master	*master;
	struct CONTROLLER	*c;

	master = spi_alloc_master(dev, sizeof *c);
	if (!master)
		return -ENODEV;

	c = class_get_devdata(&master->cdev);

The driver will initialize the fields of that spi_master, including the
bus number (maybe the same as the platform device ID) and three methods
used to interact with the SPI core and SPI protocol drivers.  It will
also initialize its own internal state.

    master->setup(struct spi_device *spi)
	This sets up the device clock rate, SPI mode, and word sizes.
	Drivers may change the defaults provided by board_info, and then
	call spi_setup(spi) to invoke this routine.  It may sleep.

    master->transfer(struct spi_device *spi, struct spi_message *message)
    	This must not sleep.  Its responsibility is arrange that the
	transfer happens and its complete() callback is issued; the two
	will normally happen later, after other transfers complete.

    master->cleanup(struct spi_device *spi)
	Your controller driver may use spi_device.controller_state to hold
	state it dynamically associates with that device.  If you do that,
	be sure to provide the cleanup() method to free that state.

The bulk of the driver will be managing the I/O queue fed by transfer().

That queue could be purely conceptual.  For example, a driver used only
for low-frequency sensor acess might be fine using synchronous PIO.

But the queue will probably be very real, using message->queue, PIO,
often DMA (especially if the root filesystem is in SPI flash), and
execution contexts like IRQ handlers, tasklets, or workqueues (such
as keventd).  Your driver can be as fancy, or as simple, as you need.


THANKS TO
---------
Contributors to Linux-SPI discussions include (in alphabetical order,
by last name):

David Brownell
Russell King
Dmitry Pervushin
Stephen Street
Mark Underwood
Andrew Victor
Vitaly Wool