Linux/AM3358: GPIO wake up from standby

Hi Matthew,

Do you use AM335x TI board (EVM, SK, BBB) or custom board? I would suggest you to first work with AM335x TI PSDK4.1 (kernel 4.9.41) only, once successful you can transfer the code to your 4.11 kernel.

Do you modify your DTS file with GPIO2 gpio-key,wakeup feature?

I will check regarding your specific question, meanwhile see if the below links will be in help:

processors.wiki.ti.com/.../Linux_Core_Power_Management_User's_Guide_(v4.4)
processors.wiki.ti.com/.../Debugging_AM335x_Suspend-Resume_Issues

www.ti.com/.../sprac74a.pdf

e2e.ti.com/.../572939
e2e.ti.com/.../550811
e2e.ti.com/.../546397
e2e.ti.com/.../498933
e2e.ti.com/.../626971

Regards,
Pavel

Matthew Harlan said:

 I have narrowed my issue to the ENAWAKEUP bit in the GPIO_SYSCONFIG.  It is not being set by the drivers in the SDK nor mainline source.

The GPIO linux driver (omap4-gpio) is:

linux-kernel/drivers/gpio/gpio-omap.c

linux-kernel/include/linux/platform_data/gpio-omap.h

I confirm that GPIO_SYSCONFIG[2] ENAWAKEUP is not modified by the GPIO driver.

This register and bit seems to be described and modified in the below files:

linux-kernel/arch/arm/mach-omap2/omap_hwmod_33xx_43xx_ipblock_data.c

/*
 * 'gpio' class: for gpio 0,1,2,3
 */
static struct omap_hwmod_class_sysconfig am33xx_gpio_sysc = {
    .rev_offs    = 0x0000,
    .sysc_offs    = 0x0010,
    .syss_offs    = 0x0114,
    .sysc_flags    = (SYSC_HAS_AUTOIDLE | SYSC_HAS_ENAWAKEUP |
              SYSC_HAS_SIDLEMODE | SYSC_HAS_SOFTRESET |
              SYSS_HAS_RESET_STATUS),
    .idlemodes    = (SIDLE_FORCE | SIDLE_NO | SIDLE_SMART |
              SIDLE_SMART_WKUP),
    .sysc_fields    = &omap_hwmod_sysc_type1,
};

linux-kernel/arch/arm/mach-omap2/omap_hwmod.h

linux-kernel/arch/arm/mach-omap2/omap_hwmod.c

Regards,
Pavel

Matthew Harlan said:

While digging into that, I am noticing there are register offsets being manipulated for wakeup enables that are not defined in the technical ref for the AM335x family.

Namely,  #define OMAP4_GPIO_WAKE_EN 0x0120

That offset is not listed in the GPIO register definition, and so far I have not been able to find who is actually setting the wakeup enable bit.

We do not have such register (at offset 0x120) in AM335x TRM, and this OMAP4_GPIO_WAKE_EN register is not used in the gpio-omap.c driver. The only wake-up related register used in the gpio-omap.c driver is OMAP4_GPIO_IRQWAKEN0 (offset 0x44).

Regards,
Pavel

Pavel,

This is on a custom board.

For my GPIO0 wakeup, i have this dtsi included in my build.

/ {

    vbatt_wakeup {
        compatible = "gpio-keys";
        pinctrl-names = "default";
        pinctrl-0 = <&vbatt_wake_pins>;

        wake@0 {
            label = "wake";
            linux,code = <KEY_WAKEUP>;
            gpios = <&gpio0 6 GPIO_ACTIVE_HIGH>;
            wakeup-source;
            }; 
    };
};

I get proper operation from this input, in both Deepsleep0 and Standby, but I know GPIO0 behaves a but differently due to the wakeup domain vs peripheral.

For my GPIO2, it is being used as the interrupt from our touch screen controller.  I have removed the touch driver and replaced with a similar device tree entry.

/ {

    LCD_wakeup {
        compatible = "gpio-keys";
        pinctrl-names = "default";
        pinctrl-0 = <&tp_gpio_pins>;

        wake@0 {
            label = "TP";
            linux,code = <KEY_WAKEUP>;
            gpios = <&gpio2 5 GPIO_ACTIVE_HIGH>;
            wakeup-source;
            };
        };
};

This does not work for GPIO2.

I have dumped the GPIO registers on suspend and resume, my IRQWAKE registers are set correctly.  The module's SYSCONFIG ENWAKEUP never seems to get set.  But if I do it manually it works.

This happens on a 4.9 SDK based kernel and my 4.11 kernel.

I have a feeling this is buried somewhere in the hwmod drivers, as I have tracked down the enable_wakeup calls within hwmod.  But as of yet I have not found where the SYSCONFIG value is setup automatically.

Matt

Matt,

Matthew Harlan said:

For my GPIO2, it is being used as the interrupt from our touch screen controller.  I have removed the touch driver and replaced with a similar device tree entry.

/ {

    LCD_wakeup {
        compatible = "gpio-keys";
        pinctrl-names = "default";
        pinctrl-0 = <&tp_gpio_pins>;

        wake@0 {
            label = "TP";
            linux,code = <KEY_WAKEUP>;
            gpios = <&gpio2 5 GPIO_ACTIVE_HIGH>;
            wakeup-source;
            };
        };
};

This does not work for GPIO2.

Your DTS file looks correct. You can also check gpio_keys.c driver, make sure wakeup-source entry is parsed (button->wakeup is set to 1), the functions device_init_wakeup(), enable_irq_wake(), pm_stay_awake(), pm_wakeup_event() are called, CONFIG_PM_SLEEP is defined.

Make sure also you have the wake-up event at the gpio2_5 pin, check with scope.

Matthew Harlan said:

I have dumped the GPIO registers on suspend and resume, my IRQWAKE registers are set correctly.  The module's SYSCONFIG ENWAKEUP never seems to get set.  But if I do it manually it works.

As you already know, this register/bit is set in omap_hwmod.c driver during boot up and during suspend/resume. Check if the related functions are called: omap_hwmod_enable(), _omap_device_enable_hwmods(), omap_device_enable(), omap_device_build_from_dt(), _omap_device_notifier_call()

_omap_device_enable_hwmods(), omap_device_enable(), omap_device_build_from_dt(), _omap_device_notifier_call() functions are located at:

linux-kernel/arch/arm/mach-omap2/omap_device.c

Regards,
Pavel