Documentation/acpi/enumeration.txt - maze/linux - Git at Google

 ACPI based device enumeration
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 ACPI 5 introduced a set of new resources (UartTSerialBus, I2cSerialBus,
 SpiSerialBus, GpioIo and GpioInt) which can be used in enumerating slave
 devices behind serial bus controllers.

 In addition we are starting to see peripherals integrated in the
 SoC/Chipset to appear only in ACPI namespace. These are typically devices
 that are accessed through memory-mapped registers.

 In order to support this and re-use the existing drivers as much as
 possible we decided to do following:

 	o Devices that have no bus connector resource are represented as
 	  platform devices.

 	o Devices behind real busses where there is a connector resource
 	  are represented as struct spi_device or struct i2c_device
 	  (standard UARTs are not busses so there is no struct uart_device).

 As both ACPI and Device Tree represent a tree of devices (and their
 resources) this implementation follows the Device Tree way as much as
 possible.

 The ACPI implementation enumerates devices behind busses (platform, SPI and
 I2C), creates the physical devices and binds them to their ACPI handle in
 the ACPI namespace.

 This means that when ACPI_HANDLE(dev) returns non-NULL the device was
 enumerated from ACPI namespace. This handle can be used to extract other
 device-specific configuration. There is an example of this below.

 Platform bus support
 ~~~~~~~~~~~~~~~~~~~~
 Since we are using platform devices to represent devices that are not
 connected to any physical bus we only need to implement a platform driver
 for the device and add supported ACPI IDs. If this same IP-block is used on
 some other non-ACPI platform, the driver might work out of the box or needs
 some minor changes.

 Adding ACPI support for an existing driver should be pretty
 straightforward. Here is the simplest example:

 	#ifdef CONFIG_ACPI
 	static struct acpi_device_id mydrv_acpi_match[] = {
 		/* ACPI IDs here */
 		{ }
 	};
 	MODULE_DEVICE_TABLE(acpi, mydrv_acpi_match);
 	#endif

 	static struct platform_driver my_driver = {
 		...
 		.driver = {
 			.acpi_match_table = ACPI_PTR(mydrv_acpi_match),
 		},
 	};

 If the driver needs to perform more complex initialization like getting and
 configuring GPIOs it can get its ACPI handle and extract this information
 from ACPI tables.

 Currently the kernel is not able to automatically determine from which ACPI
 device it should make the corresponding platform device so we need to add
 the ACPI device explicitly to acpi_platform_device_ids list defined in
 drivers/acpi/acpi_platform.c. This limitation is only for the platform
 devices, SPI and I2C devices are created automatically as described below.

 DMA support
 ~~~~~~~~~~~
 DMA controllers enumerated via ACPI should be registered in the system to
 provide generic access to their resources. For example, a driver that would
 like to be accessible to slave devices via generic API call
 dma_request_slave_channel() must register itself at the end of the probe
 function like this:

 	err = devm_acpi_dma_controller_register(dev, xlate_func, dw);
 	/* Handle the error if it's not a case of !CONFIG_ACPI */

 and implement custom xlate function if needed (usually acpi_dma_simple_xlate()
 is enough) which converts the FixedDMA resource provided by struct
 acpi_dma_spec into the corresponding DMA channel. A piece of code for that case
 could look like:

 	#ifdef CONFIG_ACPI
 	struct filter_args {
 		/* Provide necessary information for the filter_func */
 		...
 	};

 	static bool filter_func(struct dma_chan *chan, void *param)
 	{
 		/* Choose the proper channel */
 		...
 	}

 	static struct dma_chan *xlate_func(struct acpi_dma_spec *dma_spec,
 			struct acpi_dma *adma)
 	{
 		dma_cap_mask_t cap;
 		struct filter_args args;

 		/* Prepare arguments for filter_func */
 		...
 		return dma_request_channel(cap, filter_func, &args);
 	}
 	#else
 	static struct dma_chan *xlate_func(struct acpi_dma_spec *dma_spec,
 			struct acpi_dma *adma)
 	{
 		return NULL;
 	}
 	#endif

 dma_request_slave_channel() will call xlate_func() for each registered DMA
 controller. In the xlate function the proper channel must be chosen based on
 information in struct acpi_dma_spec and the properties of the controller
 provided by struct acpi_dma.

 Clients must call dma_request_slave_channel() with the string parameter that
 corresponds to a specific FixedDMA resource. By default "tx" means the first
 entry of the FixedDMA resource array, "rx" means the second entry. The table
 below shows a layout:

 	Device (I2C0)
 	{
 		...
 		Method (_CRS, 0, NotSerialized)
 		{
 			Name (DBUF, ResourceTemplate ()
 			{
 				FixedDMA (0x0018, 0x0004, Width32bit, _Y48)
 				FixedDMA (0x0019, 0x0005, Width32bit, )
 			})
 		...
 		}
 	}

 So, the FixedDMA with request line 0x0018 is "tx" and next one is "rx" in
 this example.

 In robust cases the client unfortunately needs to call
 acpi_dma_request_slave_chan_by_index() directly and therefore choose the
 specific FixedDMA resource by its index.

 SPI serial bus support
 ~~~~~~~~~~~~~~~~~~~~~~
 Slave devices behind SPI bus have SpiSerialBus resource attached to them.
 This is extracted automatically by the SPI core and the slave devices are
 enumerated once spi_register_master() is called by the bus driver.

 Here is what the ACPI namespace for a SPI slave might look like:

 	Device (EEP0)
 	{
 		Name (_ADR, 1)
 		Name (_CID, Package() {
 			"ATML0025",
 			"AT25",
 		})
 		...
 		Method (_CRS, 0, NotSerialized)
 		{
 			SPISerialBus(1, PolarityLow, FourWireMode, 8,
 				ControllerInitiated, 1000000, ClockPolarityLow,
 				ClockPhaseFirst, "\\_SB.PCI0.SPI1",)
 		}
 		...

 The SPI device drivers only need to add ACPI IDs in a similar way than with
 the platform device drivers. Below is an example where we add ACPI support
 to at25 SPI eeprom driver (this is meant for the above ACPI snippet):

 	#ifdef CONFIG_ACPI
 	static struct acpi_device_id at25_acpi_match[] = {
 		{ "AT25", 0 },
 		{ },
 	};
 	MODULE_DEVICE_TABLE(acpi, at25_acpi_match);
 	#endif

 	static struct spi_driver at25_driver = {
 		.driver = {
 			...
 			.acpi_match_table = ACPI_PTR(at25_acpi_match),
 		},
 	};

 Note that this driver actually needs more information like page size of the
 eeprom etc. but at the time writing this there is no standard way of
 passing those. One idea is to return this in _DSM method like:

 	Device (EEP0)
 	{
 		...
 		Method (_DSM, 4, NotSerialized)
 		{
 			Store (Package (6)
 			{
 				"byte-len", 1024,
 				"addr-mode", 2,
 				"page-size, 32
 			}, Local0)

 			// Check UUIDs etc.

 			Return (Local0)
 		}

 Then the at25 SPI driver can get this configuration by calling _DSM on its
 ACPI handle like:

 	struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL };
 	struct acpi_object_list input;
 	acpi_status status;

 	/* Fill in the input buffer */

 	status = acpi_evaluate_object(ACPI_HANDLE(&spi->dev), "_DSM",
 				      &input, &output);
 	if (ACPI_FAILURE(status))
 		/* Handle the error */

 	/* Extract the data here */

 	kfree(output.pointer);

 I2C serial bus support
 ~~~~~~~~~~~~~~~~~~~~~~
 The slaves behind I2C bus controller only need to add the ACPI IDs like
 with the platform and SPI drivers. The I2C core automatically enumerates
 any slave devices behind the controller device once the adapter is
 registered.

 Below is an example of how to add ACPI support to the existing mpu3050
 input driver:

 	#ifdef CONFIG_ACPI
 	static struct acpi_device_id mpu3050_acpi_match[] = {
 		{ "MPU3050", 0 },
 		{ },
 	};
 	MODULE_DEVICE_TABLE(acpi, mpu3050_acpi_match);
 	#endif

 	static struct i2c_driver mpu3050_i2c_driver = {
 		.driver	= {
 			.name	= "mpu3050",
 			.owner	= THIS_MODULE,
 			.pm	= &mpu3050_pm,
 			.of_match_table = mpu3050_of_match,
 			.acpi_match_table  ACPI_PTR(mpu3050_acpi_match),
 		},
 		.probe		= mpu3050_probe,
 		.remove		= mpu3050_remove,
 		.id_table	= mpu3050_ids,
 	};

 GPIO support
 ~~~~~~~~~~~~
 ACPI 5 introduced two new resources to describe GPIO connections: GpioIo
 and GpioInt. These resources are used be used to pass GPIO numbers used by
 the device to the driver. For example:

 	Method (_CRS, 0, NotSerialized)
 	{
 		Name (SBUF, ResourceTemplate()
 		{
 			...
 			// Used to power on/off the device
 			GpioIo (Exclusive, PullDefault, 0x0000, 0x0000,
 				IoRestrictionOutputOnly, "\\_SB.PCI0.GPI0",
 				0x00, ResourceConsumer,,)
 			{
 				// Pin List
 				0x0055
 			}

 			// Interrupt for the device
 			GpioInt (Edge, ActiveHigh, ExclusiveAndWake, PullNone,
 				 0x0000, "\\_SB.PCI0.GPI0", 0x00, ResourceConsumer,,)
 			{
 				// Pin list
 				0x0058
 			}

 			...

 		}

 		Return (SBUF)
 	}

 These GPIO numbers are controller relative and path "\\_SB.PCI0.GPI0"
 specifies the path to the controller. In order to use these GPIOs in Linux
 we need to translate them to the corresponding Linux GPIO descriptors.

 There is a standard GPIO API for that and is documented in
 Documentation/gpio.txt.

 In the above example we can get the corresponding two GPIO descriptors with
 a code like this:

 	#include <linux/gpio/consumer.h>
 	...

 	struct gpio_desc *irq_desc, *power_desc;

 	irq_desc = gpiod_get_index(dev, NULL, 1);
 	if (IS_ERR(irq_desc))
 		/* handle error */

 	power_desc = gpiod_get_index(dev, NULL, 0);
 	if (IS_ERR(power_desc))
 		/* handle error */

 	/* Now we can use the GPIO descriptors */

 There are also devm_* versions of these functions which release the
 descriptors once the device is released.
	ACPI based device enumeration
	~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
	ACPI 5 introduced a set of new resources (UartTSerialBus, I2cSerialBus,
	SpiSerialBus, GpioIo and GpioInt) which can be used in enumerating slave
	devices behind serial bus controllers.

	In addition we are starting to see peripherals integrated in the
	SoC/Chipset to appear only in ACPI namespace. These are typically devices
	that are accessed through memory-mapped registers.

	In order to support this and re-use the existing drivers as much as
	possible we decided to do following:

	o Devices that have no bus connector resource are represented as
	platform devices.

	o Devices behind real busses where there is a connector resource
	are represented as struct spi_device or struct i2c_device
	(standard UARTs are not busses so there is no struct uart_device).

	As both ACPI and Device Tree represent a tree of devices (and their
	resources) this implementation follows the Device Tree way as much as
	possible.

	The ACPI implementation enumerates devices behind busses (platform, SPI and
	I2C), creates the physical devices and binds them to their ACPI handle in
	the ACPI namespace.

	This means that when ACPI_HANDLE(dev) returns non-NULL the device was
	enumerated from ACPI namespace. This handle can be used to extract other
	device-specific configuration. There is an example of this below.

	Platform bus support
	~~~~~~~~~~~~~~~~~~~~
	Since we are using platform devices to represent devices that are not
	connected to any physical bus we only need to implement a platform driver
	for the device and add supported ACPI IDs. If this same IP-block is used on
	some other non-ACPI platform, the driver might work out of the box or needs
	some minor changes.

	Adding ACPI support for an existing driver should be pretty
	straightforward. Here is the simplest example:

	#ifdef CONFIG_ACPI
	static struct acpi_device_id mydrv_acpi_match[] = {
	/* ACPI IDs here */
	{ }
	};
	MODULE_DEVICE_TABLE(acpi, mydrv_acpi_match);
	#endif

	static struct platform_driver my_driver = {
	...
	.driver = {
	.acpi_match_table = ACPI_PTR(mydrv_acpi_match),
	},
	};

	If the driver needs to perform more complex initialization like getting and
	configuring GPIOs it can get its ACPI handle and extract this information
	from ACPI tables.

	Currently the kernel is not able to automatically determine from which ACPI
	device it should make the corresponding platform device so we need to add
	the ACPI device explicitly to acpi_platform_device_ids list defined in
	drivers/acpi/acpi_platform.c. This limitation is only for the platform
	devices, SPI and I2C devices are created automatically as described below.

	DMA support
	~~~~~~~~~~~
	DMA controllers enumerated via ACPI should be registered in the system to
	provide generic access to their resources. For example, a driver that would
	like to be accessible to slave devices via generic API call
	dma_request_slave_channel() must register itself at the end of the probe
	function like this:

	err = devm_acpi_dma_controller_register(dev, xlate_func, dw);
	/* Handle the error if it's not a case of !CONFIG_ACPI */

	and implement custom xlate function if needed (usually acpi_dma_simple_xlate()
	is enough) which converts the FixedDMA resource provided by struct
	acpi_dma_spec into the corresponding DMA channel. A piece of code for that case
	could look like:

	#ifdef CONFIG_ACPI
	struct filter_args {
	/* Provide necessary information for the filter_func */
	...
	};

	static bool filter_func(struct dma_chan chan, void param)
	{
	/* Choose the proper channel */
	...
	}

	static struct dma_chan xlate_func(struct acpi_dma_spec dma_spec,
	struct acpi_dma *adma)
	{
	dma_cap_mask_t cap;
	struct filter_args args;

	/* Prepare arguments for filter_func */
	...
	return dma_request_channel(cap, filter_func, &args);
	}
	#else
	static struct dma_chan xlate_func(struct acpi_dma_spec dma_spec,
	struct acpi_dma *adma)
	{
	return NULL;
	}
	#endif

	dma_request_slave_channel() will call xlate_func() for each registered DMA
	controller. In the xlate function the proper channel must be chosen based on
	information in struct acpi_dma_spec and the properties of the controller
	provided by struct acpi_dma.

	Clients must call dma_request_slave_channel() with the string parameter that
	corresponds to a specific FixedDMA resource. By default "tx" means the first
	entry of the FixedDMA resource array, "rx" means the second entry. The table
	below shows a layout:

	Device (I2C0)
	{
	...
	Method (_CRS, 0, NotSerialized)
	{
	Name (DBUF, ResourceTemplate ()
	{
	FixedDMA (0x0018, 0x0004, Width32bit, _Y48)
	FixedDMA (0x0019, 0x0005, Width32bit, )
	})
	...
	}
	}

	So, the FixedDMA with request line 0x0018 is "tx" and next one is "rx" in
	this example.

	In robust cases the client unfortunately needs to call
	acpi_dma_request_slave_chan_by_index() directly and therefore choose the
	specific FixedDMA resource by its index.

	SPI serial bus support
	~~~~~~~~~~~~~~~~~~~~~~
	Slave devices behind SPI bus have SpiSerialBus resource attached to them.
	This is extracted automatically by the SPI core and the slave devices are
	enumerated once spi_register_master() is called by the bus driver.

	Here is what the ACPI namespace for a SPI slave might look like:

	Device (EEP0)
	{
	Name (_ADR, 1)
	Name (_CID, Package() {
	"ATML0025",
	"AT25",
	})
	...
	Method (_CRS, 0, NotSerialized)
	{
	SPISerialBus(1, PolarityLow, FourWireMode, 8,
	ControllerInitiated, 1000000, ClockPolarityLow,
	ClockPhaseFirst, "\\_SB.PCI0.SPI1",)
	}
	...

	The SPI device drivers only need to add ACPI IDs in a similar way than with
	the platform device drivers. Below is an example where we add ACPI support
	to at25 SPI eeprom driver (this is meant for the above ACPI snippet):

	#ifdef CONFIG_ACPI
	static struct acpi_device_id at25_acpi_match[] = {
	{ "AT25", 0 },
	{ },
	};
	MODULE_DEVICE_TABLE(acpi, at25_acpi_match);
	#endif

	static struct spi_driver at25_driver = {
	.driver = {
	...
	.acpi_match_table = ACPI_PTR(at25_acpi_match),
	},
	};

	Note that this driver actually needs more information like page size of the
	eeprom etc. but at the time writing this there is no standard way of
	passing those. One idea is to return this in _DSM method like:

	Device (EEP0)
	{
	...
	Method (_DSM, 4, NotSerialized)
	{
	Store (Package (6)
	{
	"byte-len", 1024,
	"addr-mode", 2,
	"page-size, 32
	}, Local0)

	// Check UUIDs etc.

	Return (Local0)
	}

	Then the at25 SPI driver can get this configuration by calling _DSM on its
	ACPI handle like:

	struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL };
	struct acpi_object_list input;
	acpi_status status;

	/* Fill in the input buffer */

	status = acpi_evaluate_object(ACPI_HANDLE(&spi->dev), "_DSM",
	&input, &output);
	if (ACPI_FAILURE(status))
	/* Handle the error */

	/* Extract the data here */

	kfree(output.pointer);

	I2C serial bus support
	~~~~~~~~~~~~~~~~~~~~~~
	The slaves behind I2C bus controller only need to add the ACPI IDs like
	with the platform and SPI drivers. The I2C core automatically enumerates
	any slave devices behind the controller device once the adapter is
	registered.

	Below is an example of how to add ACPI support to the existing mpu3050
	input driver:

	#ifdef CONFIG_ACPI
	static struct acpi_device_id mpu3050_acpi_match[] = {
	{ "MPU3050", 0 },
	{ },
	};
	MODULE_DEVICE_TABLE(acpi, mpu3050_acpi_match);
	#endif

	static struct i2c_driver mpu3050_i2c_driver = {
	.driver = {
	.name = "mpu3050",
	.owner = THIS_MODULE,
	.pm = &mpu3050_pm,
	.of_match_table = mpu3050_of_match,
	.acpi_match_table ACPI_PTR(mpu3050_acpi_match),
	},
	.probe = mpu3050_probe,
	.remove = mpu3050_remove,
	.id_table = mpu3050_ids,
	};

	GPIO support
	~~~~~~~~~~~~
	ACPI 5 introduced two new resources to describe GPIO connections: GpioIo
	and GpioInt. These resources are used be used to pass GPIO numbers used by
	the device to the driver. For example:

	Method (_CRS, 0, NotSerialized)
	{
	Name (SBUF, ResourceTemplate()
	{
	...
	// Used to power on/off the device
	GpioIo (Exclusive, PullDefault, 0x0000, 0x0000,
	IoRestrictionOutputOnly, "\\_SB.PCI0.GPI0",
	0x00, ResourceConsumer,,)
	{
	// Pin List
	0x0055
	}

	// Interrupt for the device
	GpioInt (Edge, ActiveHigh, ExclusiveAndWake, PullNone,
	0x0000, "\\_SB.PCI0.GPI0", 0x00, ResourceConsumer,,)
	{
	// Pin list
	0x0058
	}

	...

	}

	Return (SBUF)
	}

	These GPIO numbers are controller relative and path "\\_SB.PCI0.GPI0"
	specifies the path to the controller. In order to use these GPIOs in Linux
	we need to translate them to the corresponding Linux GPIO descriptors.

	There is a standard GPIO API for that and is documented in
	Documentation/gpio.txt.

	In the above example we can get the corresponding two GPIO descriptors with
	a code like this:

	#include <linux/gpio/consumer.h>
	...

	struct gpio_desc irq_desc, power_desc;

	irq_desc = gpiod_get_index(dev, NULL, 1);
	if (IS_ERR(irq_desc))
	/* handle error */

	power_desc = gpiod_get_index(dev, NULL, 0);
	if (IS_ERR(power_desc))
	/* handle error */

	/* Now we can use the GPIO descriptors */

	There are also devm_* versions of these functions which release the
	descriptors once the device is released.