Let's take a look at an example entry from src/mainboard/google/hatch/variants/hatch/overridetree.cb
:
device pci 15.0 on chip drivers/i2c/generic register "hid" = ""ELAN0000"" register "desc" = ""ELAN Touchpad"" register "irq" = "ACPI_IRQ_LEVEL_LOW(GPP_A21_IRQ)" register "detect" = "1" register "wake" = "GPE0_DW0_21" device i2c 15 on end end end # I2C #0
When this entry is processed during ramstage, it will create a device in the ACPI SSDT table (all devices in devicetrees end up in the SSDT table). The ACPI generation routines in coreboot actually generate the raw bytecode that represents the device's structure, but looking at ASL code is easier to understand; see below for what the disassembled bytecode looks like:
Scope (\_SB.PCI0.I2C0) { Device (D015) { Name (_HID, "ELAN0000") // _HID: Hardware ID Name (_UID, Zero) // _UID: Unique ID Name (_DDN, "ELAN Touchpad") // _DDN: DOS Device Name Method (_STA, 0, NotSerialized) // _STA: Status { Return (0x0F) } Name (_CRS, ResourceTemplate () // _CRS: Current Resource Settings { I2cSerialBusV2 (0x0015, ControllerInitiated, 400000, AddressingMode7Bit, "\\_SB.PCI0.I2C0", 0x00, ResourceConsumer, , Exclusive, ) Interrupt (ResourceConsumer, Level, ActiveLow, Exclusive, ,, ) { 0x0000002D, } }) Name (_S0W, ACPI_DEVICE_SLEEP_D3_HOT) // _S0W: S0 Device Wake State Name (_PRW, Package (0x02) // _PRW: Power Resources for Wake { 0x15, // GPE #21 0x03 // Sleep state S3 }) } }
You can see it generates _HID, _UID, _DDN, _STA, _CRS, _S0W, and _PRW names/methods in the Device's scope.
The device driver must be enabled for your build. There will be a CONFIG option in the Kconfig file in the directory that the driver is in (e.g., src/drivers/i2c/generic
contains a Kconfig file; the option here is named CONFIG_DRIVERS_I2C_GENERIC). The config option will need to be added to your mainboard's Kconfig file (e.g., src/mainboard/google/hatch/Kconfig
) in order to be compiled into your build.
Let's take a look at how the devicetree language corresponds to the generated ASL.
First, note this:
chip drivers/i2c/generic
This means that the device driver we're using has a corresponding structure, located at src/drivers/i2c/generic/chip.h
, named struct drivers_i2c_generic_config and it contains many properties you can specify to be included in the ACPI table.
register "hid" = ""ELAN0000""
This corresponds to const char *hid in the struct. In the ACPI ASL, it translates to:
Name (_HID, "ELAN0000") // _HID: Hardware ID
under the device. This property is used to match the device to its driver during enumeration in the OS.
register "desc" = ""ELAN Touchpad""
corresponds to const char *desc and in ASL:
Name (_DDN, "ELAN Touchpad") // _DDN: DOS Device Name
It also adds the interrupt,
Interrupt (ResourceConsumer, Level, ActiveLow, Exclusive, ,, ) { 0x0000002D, }
which comes from:
register "irq" = "ACPI_IRQ_LEVEL_LOW(GPP_A21_IRQ)"
The IRQ settings control the "Trigger" and "Polarity" settings seen above (level means it is a level-triggered interrupt as opposed to edge-triggered; active low means the interrupt is triggered when the signal is low).
Also note that the IRQ names are SoC-specific, and you will need to find the names in your SoC's header file. The ACPI_* macros are defined in src/arch/x86/include/acpi/acpi_device.h
.
Using a GPIO as an IRQ requires that it is configured in coreboot correctly. This is often done in a mainboard-specific file named gpio.c
.
AMD platforms don't have the ability to route GPIOs to the IO-APIC. Instead the GPIO controller needs to be used directly. You can do this by setting the irq_gpio
register and using the ACPI_GPIO_IRQ_X_X
macros.
i.e.,
register "irq_gpio" = "ACPI_GPIO_IRQ_EDGE_LOW(GPIO_40)"
The next register is:
register "detect" = "1"
This flag tells the I2C driver that it should attempt to detect the presence of the device (using an I2C zero-byte write), and only generate a SSDT entry if the device is actually present. This alleviates the OS from having to determine if a device is present or not (ChromeOS/Linux) and prevents resource conflict/ driver issues (Windows).
Currently, the detect feature works and is hooked up for all I2C touchpads, and should be used any time a board has multiple touchpad options. I2C audio devices should also work without issue.
Touchscreens can use this feature as well, but special care is needed to implement the proper power sequencing for the device to be detected. Generally, this means driving the enable GPIO high and holding the reset GPIO low in early GPIO init (bootblock/romstage), then releasing reset in ramstage. The first mainboards in the tree to implement this are google/skyrim and google/guybrush. This feature has also been used in downstream forks without issue for some time now on several other boards.
The last register is:
register "wake" = "GPE0_DW0_21"
which indicates that the method of waking the system using the touchpad will be through a GPE, #21 associated with DW0, which is set up in devicetree.cb from this example. The "21" indicates GPP_X21, where GPP_X is mapped onto DW0 elsewhere in the devicetree.
The last bit of the definition of that device includes:
device i2c 15 on end
which means it's an I2C device, with 7-bit address 0x15, and the device is "on", meaning it will be exposed in the ACPI table. The PCI device that the controller is located in determines which I2C bus the device is expected to be found on. In this example, this is I2C bus 0. This also determines the ACPI "Scope" that the device names and methods will live under, in this case "_SB.PCI0.I2C0".
The ACPI spec defines two methods to describe how a device can wake the system. Only one of these methods should be used, otherwise duplicate wake events will be generated.
The wake
property specified above is used to tell the ACPI subsystem that the device can use a GPE to wake the system. The OS can control whether to enable or disable the wake source by unmasking/masking off the GPE.
The GPIO
-> GPE
mapping must be configured in firmware. On AMD platforms this is generally done by a mainboard specific gpio.c
file that defines the GPIO using PAD_SCI
. The GPIO
-> GPE
mapping is returned by the soc_get_gpio_event_table
method that is defined in the SoC specific gpio.c
file. On Intel platforms, you fill in the pmc_gpe0_dw0
, pmc_gpe0_dw1
, and pmc_gpe0_dw2
fields in the devicetree to map 3 GPIO communities to tier-1
GPEs (the rest are available as tier-2
GPEs).
Windows has a large caveat when using this method. If you use the gpio_irq
property to define a GpioInt
in the _CRS
, and then use the wake
property to define a GPE
, Windows will BSOD complaining about an invalid ACPI configuration.
0x1000D - A device used both GPE and GPIO interrupts, which is not supported.
In order to avoid this error, you should use the irq
property instead. AMD platforms don't support routing GPIOs to the IO-APIC, so this workaround isn't feasible. The other option is to use a wake capable GPIO as described below.
The ACPI_IRQ_WAKE_{EDGE,LEVEL}_{LOW,HIGH}
macros can be used when setting the irq
or gpio_irq
properties. This ends up setting ExclusiveAndWake
or SharedAndWake
on the Interrupt
or GpioInt
ACPI resource.
This method has a few caveats:
ACPI_IRQ_WAKE_*
macros with the irq
property won't actually wake the system. Instead you need to use the gpio_irq
property, or a GPE
as described above.wake
bit on the GPIO. For linux this means the platform specific GPIO controller driver must implement the irq_set_wake
callback. For AMD systems this wasn't implemented until linux v5.15. If the controller doesn't define this callback, it's possible for the firmware to manually set the wake
bit on the GPIO. This is often done in a mainboard-specific file named gpio.c
. This is not recommended because then it's not possible for the OS to disable the wake source.wake
bit into account when deciding on which power state to put the device in before suspending the system. This means that if you define a power resource for a device via has_power_resource
, enable_gpio
, etc, then the linux kernel will place the device into D3Cold. i.e., power off the device.(see ACPI specification 6.3 for more details on ACPI methods)
_S0W indicates the deepest S0 sleep state this device can wake itself from, which in this case is ACPI_DEVICE_SLEEP_D3_HOT
, representing D3hot. D3Hot means the PR3
power resources are still on and the device is still responsive on the bus. For i2c devices this is generally the same state as D0
.
_PRW indicates the power resources and events required for wake. There are no dependent power resources, but the GPE (GPE0_DW0_21) is mentioned here (0x15), as well as the deepest sleep state supporting waking the system (3), which is S3.
The _STA method is generated automatically, and its values, 0xF, indicates the following:
Bit [0] – Set if the device is present. Bit [1] – Set if the device is enabled and decoding its resources. Bit [2] – Set if the device should be shown in the UI. Bit [3] – Set if the device is functioning properly (cleared if device failed its diagnostics).
The _CRS method is generated automatically, as the driver knows it is an I2C controller, and so specifies how to configure the controller for proper operation with the touchpad.
Name (_CRS, ResourceTemplate () // _CRS: Current Resource Settings { I2cSerialBusV2 (0x0015, ControllerInitiated, 400000, AddressingMode7Bit, "\\_SB.PCI0.I2C0", 0x00, ResourceConsumer, , Exclusive, )