Configuring a mainboard’s GPIOs in coreboot

Introduction

Every mainboard needs to appropriately configure its General Purpose Inputs / Outputs (GPIOs). There are many facets of this issue, including which boot stage a GPIO might need to be configured.

Boot stages

Typically, coreboot does most of its non-memory related initialization work in ramstage, when DRAM is available for use. Hence, the bulk of a mainboard’s GPIOs are configured in this stage. However, some boards might need a few GPIOs configured before that; think of memory strapping pins which indicate what kind of DRAM is installed. These pins might need to be read before initializing the memory, so these GPIOs are then typically configured in bootblock or romstage.

Configuration

Most mainboards will have a gpio.c file in their mainboard directory. This file typically contains tables which describe the configuration of the GPIO registers. Since these registers could be different on a per-SoC or per SoC-family basis, you may need to consult the datasheet for your SoC to find out how to appropriately set these registers. In addition, some mainboards are based on a baseboard/variant model, where several variant mainboards may share a lot of their circuitry and ICs and the commonality between the boards is collected into a virtual baseboard. In that case, the GPIOs which are shared between multiple boards are placed in the baseboard’s gpio.c file, while the ones that are board-specific go into each variant’s gpio.c file.

Intel SoCs

Many newer Intel SoCs share a common IP block for GPIOs, and that commonality has been taken advantage of in coreboot, which has a large set of macros that can be used to describe the configuration of each GPIO pad. This file lives in src/soc/intel/common/block/include/intelblocks/gpio_defs.h.

Older Intel SoCs

Baytrail and Braswell, for example, simply expect the mainboard to supply a callback, mainboard_get_gpios which returns an array of struct soc_gpio objects, defining the configuration of each pin.

AMD SoCs

Some AMD SoCs use a list of struct soc_amd_gpio objects to define the register values configuring each pin, similar to Intel.

Register details

GPIO configuration registers typically control properties such as:

  1. Input / Output
  2. Pullups / Pulldowns
  3. Termination
  4. Tx / Rx Disable
  5. Which reset signal to use
  6. Native Function / IO
  7. Interrupts
    • IRQ routing (e.g. on x86, APIC, SCI, SMI)
    • Edge or Level Triggered
    • Active High or Active Low
  8. Debouncing

Configuring GPIOs for pre-ramstage

coreboot provides for several SoC-specific and mainboard-specific callbacks at specific points in time, such as bootblock-early, bootblock, romstage entry, pre-silicon init, pre-RAM init, or post-RAM init. The GPIOs that are configured in either bootblock or romstage, depending on when they are needed, are denoted the “early” GPIOs. Some mainboard will use bootblock_mainboard_init() to configure their early GPIOs, and this is probably a good place to start. Many mainboards will declare their GPIO configuration as structs, i.e. (Intel),

struct pad_config {
    /* offset of pad within community */
        int             pad;
    /* Pad config data corresponding to DW0, DW1,.... */
        uint32_t        pad_config[GPIO_NUM_PAD_CFG_REGS];
};

and will usually place these in an array, one for each pad to be configured. Mainboards using Intel SoCs can use a library which combines common configurations together into a set of macros, e.g.,

    /* Native function configuration */
    #define PAD_CFG_NF(pad, pull, rst, func)
    /* General purpose output, no pullup/down. */
    #define PAD_CFG_GPO(pad, val, rst)
    /* General purpose output, with termination specified */
    #define PAD_CFG_TERM_GPO(pad, val, pull, rst)
    /* General purpose output, no pullup/down. */
    #define PAD_CFG_GPO_GPIO_DRIVER(pad, val, rst, pull)
    /* General purpose input */
    #define PAD_CFG_GPI(pad, pull, rst)

etc.

Configuring GPIOs for ramstage and beyond…

In ramstage, most mainboards will configure the rest of their GPIOs for the function they will be performing while the device is active. The goal is the same as above in bootblock; another static const array is created, and the rest of the GPIO registers are programmed.

In the baseboard/variant model described above, the baseboard will provide the configuration for the GPIOs which are configured identically between variants, and will provide a mechanism for a variant to override the baseboard’s configuration. This is usually done via two tables: the baseboard table and the variant’s override table.

This configuration is often hooked into the mainboard’s enable_dev callback, defined in its struct chip_operations.

Unconnected and unused pads

In digital electronics, it is generally recommended to tie unconnected GPIOs to a defined signal like VCC or GND by setting their direction to output, adding an external pull resistor or configuring an internal pull resistor. This is done to prevent floating of the pin state, which can cause various issues like EMI, higher power usage due to continuously switching logic, etc.

On Intel PCHs from Sunrise Point onwards, termination of unconnected GPIOs is explicitly not required, when the input buffer is disabled by setting the bit GPIORXDIS which effectively disconnects the pad from the internal logic. All pads defaulting to GPIO mode have this bit set. However, in the mainboard’s GPIO configuration the macro PAD_NC(pad, NONE) can be used to explicitly configure a pad as unconnected.

In case there are no schematics available for a board and the vendor set a pad to something like GPIORXDIS=1, GPIOTXDIS=1 with an internal pull resistor, an unconnected or otherwise unused pad can be assumed. In this case it is recommended to keep the pull resistor, because the external circuit might rely on it.

Unconnected pads defaulting to a native function (input and output) usually don’t need to be configured as GPIO with the GPIORXDIS bit set. For clarity and documentation purpose the macro may be used as well for them.

Some pads configured as native input function explicitly require external pull-ups when being unused, according to the PDGs:

  • eDP_HPD
  • SMBCLK/SMBDATA
  • SML0CLK/SML0DATA/SML0ALERT
  • SATAGP*

When the board was designed correctly, nothing needs to be done for them explicitly, while using PAD_NC(pad, NONE) can act as documentation. If such a pad is missing the external pull resistor due to bad board design, the pad should be configured with PAD_NC(pad, NONE) anyway to disconnect it internally.

Potential issues (gotchas!)

There are a couple of configurations that you need to especially careful about, as they can have a large impact on your mainboard.

The first is configuring a pin as an output, when it was designed to be an input. There is a real risk in this case of short-circuiting a component which could cause catastrophic failures, up to and including your mainboard!

Soft Straps

Soft straps, that can be configured by the vendor in the Intel Flash Image Tool (FIT), can influence some pads’ default mode. It is possible to select either a native function or GPIO mode for some pads on non-server SoCs, while on server SoCs most pads can be controlled. Thus, it is generally recommended to always configure all pads and don’t just rely on the defaults mentioned in the datasheet(s) which might not reflect what the vendor configured.