From 0e9cc44d1b73ee19c20b73e3bd21c7daa26f2a56 Mon Sep 17 00:00:00 2001 From: Raul E Rangel Date: Tue, 6 Sep 2022 14:32:59 -0600 Subject: Documentation: Add wake source info to device tree documentation The device tree documentation was promoting using a GPIO wake event and a GPE wake event. We should only ever have one. This wasn't actually causing a problem because the wake bit was set on the `irq` property, but the IO-APIC can't actually wake the system, so it was a no-op. This change fixes up the markdown so it's formatted correctly, and also adds a section explaining what the different wake configurations are. BUG=b:243700486 TEST=mdformat Signed-off-by: Raul E Rangel Change-Id: Ifcdbd5371408784bf9b81c1ade90263de8c60e0f Reviewed-on: https://review.coreboot.org/c/coreboot/+/67385 Reviewed-by: Karthik Ramasubramanian Reviewed-by: Tim Van Patten Tested-by: build bot (Jenkins) --- Documentation/drivers/dt_entries.md | 114 +++++++++++++++++++++++++++++------- 1 file changed, 92 insertions(+), 22 deletions(-) (limited to 'Documentation/drivers') diff --git a/Documentation/drivers/dt_entries.md b/Documentation/drivers/dt_entries.md index 9df965fa0d0e..dd54846e951b 100644 --- a/Documentation/drivers/dt_entries.md +++ b/Documentation/drivers/dt_entries.md @@ -8,7 +8,7 @@ device pci 15.0 on chip drivers/i2c/generic register "hid" = ""ELAN0000"" register "desc" = ""ELAN Touchpad"" - register "irq" = "ACPI_IRQ_WAKE_LEVEL_LOW(GPP_A21_IRQ)" + register "irq" = "ACPI_IRQ_LEVEL_LOW(GPP_A21_IRQ)" register "detect" = "1" register "wake" = "GPE0_DW0_21" device i2c 15 on end @@ -39,7 +39,7 @@ Scope (\_SB.PCI0.I2C0) I2cSerialBusV2 (0x0015, ControllerInitiated, 400000, AddressingMode7Bit, "\\_SB.PCI0.I2C0", 0x00, ResourceConsumer, , Exclusive, ) - Interrupt (ResourceConsumer, Level, ActiveLow, ExclusiveAndWake, ,, ) + Interrupt (ResourceConsumer, Level, ActiveLow, Exclusive, ,, ) { 0x0000002D, } @@ -54,7 +54,7 @@ Scope (\_SB.PCI0.I2C0) } ``` -You can see it generates _HID, _UID, _DDN, _STA, _CRS, _S0W, and _PRW +You can see it generates \_HID, \_UID, \_DDN, \_STA, \_CRS, \_S0W, and \_PRW names/methods in the Device's scope. ## Utilizing a device driver @@ -88,7 +88,7 @@ be included in the ACPI table. register "hid" = ""ELAN0000"" ``` -This corresponds to **const char *hid** in the struct. In the ACPI ASL, it +This corresponds to **const char \*hid** in the struct. In the ACPI ASL, it translates to: ``` @@ -104,7 +104,7 @@ during enumeration in the OS.** register "desc" = ""ELAN Touchpad"" ``` -corresponds to **const char *desc** and in ASL: +corresponds to **const char \*desc** and in ASL: ``` Name (_DDN, "ELAN Touchpad") // _DDN: DOS Device Name @@ -115,7 +115,7 @@ corresponds to **const char *desc** and in ASL: It also adds the interrupt, ``` - Interrupt (ResourceConsumer, Level, ActiveLow, ExclusiveAndWake, ,, ) + Interrupt (ResourceConsumer, Level, ActiveLow, Exclusive, ,, ) { 0x0000002D, } @@ -124,23 +124,30 @@ It also adds the interrupt, which comes from: ``` - register "irq" = "ACPI_IRQ_WAKE_LEVEL_LOW(GPP_A21_IRQ)" + register "irq" = "ACPI_IRQ_LEVEL_LOW(GPP_A21_IRQ)" ``` -The GPIO pin IRQ settings control the "Level", "ActiveLow", and -"ExclusiveAndWake" 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). +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). -Note that the ACPI_IRQ_WAKE_LEVEL_LOW macro informs the platform that the GPIO -will be routed through SCI (ACPI's System Control Interrupt) for use as a wake -source. Also note that the IRQ names are SoC-specific, and you will need to +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)" +``` + ### detect The next register is: @@ -180,6 +187,8 @@ 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. +### device + The last bit of the definition of that device includes: ``` @@ -193,6 +202,65 @@ 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". +## Wake sources + +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. + +### Using GPEs as a wake source + +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](https://github.com/MicrosoftDocs/windows-driver-docs/blob/staging/windows-driver-docs-pr/debugger/bug-check-0xa5--acpi-bios-error.md) +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. + +### Using GPIO interrupts as a wake source + +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: +* On Intel and AMD platforms the IO-APIC can't wake the system. This means using + the `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. +* The OS needs to know how to enable the `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](https://github.com/torvalds/linux/commit/d62bd5ce12d79bcd6a6c3e4381daa7375dc21158) + 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. +* As of + [linux v6.0-rc5](https://github.com/torvalds/linux/releases/tag/v6.0-rc5), + the ACPI subsystem doesn't take the interrupt `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. + ## Other auto-generated names (see [ACPI specification @@ -200,17 +268,19 @@ found on. In this example, this is I2C bus 0. This also determines the ACPI for more details on ACPI methods) ### _S0W (S0 Device Wake State) -_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_. +\_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 (Power Resources for Wake) -_PRW indicates the power resources and events required for wake. There are no +### \_PRW (Power Resources for Wake) +\_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. -### _STA (Status) -The _STA method is generated automatically, and its values, 0xF, indicates the +### \_STA (Status) +The \_STA method is generated automatically, and its values, 0xF, indicates the following: Bit [0] – Set if the device is present. @@ -218,8 +288,8 @@ following: 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). -### _CRS (Current resource settings) -The _CRS method is generated automatically, as the driver knows it is an I2C +### \_CRS (Current resource settings) +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. -- cgit v1.2.3