Re: [BUG] 36-second boot delay due to by acpi_gpio_handle_deferred_request_irqs on ASUS ROG Strix G16 (2025)

[Armin Wolf <W_Armin@gmx.de>]

Am 28.04.26 um 09:59 schrieb Mika Westerberg:

> Hi,
>
> On Tue, Apr 28, 2026 at 12:09:10AM +0200, Armin Wolf wrote:
>> Am 27.04.26 um 06:57 schrieb Mika Westerberg:
>>
>>> On Sat, Apr 25, 2026 at 10:41:46PM +0200, Armin Wolf wrote:
>>>>>> according to the Microsoft documentation
>>>>>> (https://learn.microsoft.com/
>>>>>> en-us/windows-hardware/drivers/bringup/general-purpose-i-o--gpio-,
>>>>>> section "GPIO controllers and ActiveBoth interrupts"), triggering
>>>>>> GPIO interrupts marked as ActiveBoth during initialization is
>>>>>> correct as long as the associated GPIO line is already "asserted"
>>>>>> (aka logic level low). I think the problem is that we also trigger
>>>>>> edge-based GPIO interrupts _not_ marked as ActiveBoth.
>>>>>>
>>>>>> Based on this i agree with Hans, except that we should continue you
>>>>>> trigger ActiveBoth GPIO interrupts as long as the above
>>>>>> condition applies.
>>>>>>
>>>>>> Thanks,
>>>>>> Armin Wolf
>>>>>>
>>>>> So maybe something like this (attached)?
>>>> Yes, exactly.
>>> This is good information and definitely scales better than the quirk list.
>>> The linked document also mentions that there is _DSM under GPIO device that
>>> could be used to override the asserted state:
>>>
>>> https://learn.microsoft.com/en-us/windows-hardware/drivers/bringup/gpio-controller-device-specific-method---dsm-
>>>
>>> I wonder if we should implement that as well?
>> I already did some work in this area, but i kind of forgot about it
>> because i have no suitable device for testing this.  Maybe someone has a
>> device that uses this _DSM and can check if the attached patch works?
> I don't have devices with the _DSM but few comments on the code (thanks BTW
> for doing this!).
>
>> Thanks,
>> Armin Wolf
>>  From 2e1dc51c864802c05c3a9e6b745dd8c34b4b615b Mon Sep 17 00:00:00 2001
>> From: Armin Wolf <W_Armin@gmx.de>
>> Date: Thu, 5 Feb 2026 22:39:35 +0100
>> Subject: [PATCH] gpiolib: acpi: Add support for Microsoft GPIO _DSM
>>
>> Microsoft has defined a ACPI device specific method for determining
>> the logic level for which a given ActiveBoth GPIO interrupt should be
>> considered "asserted". Such GPIO interrupts are normally considered to
>> be "asserted" when being logic level low.
>> Being able to override this assumption might be important for firmware
>> implementations where a ActiveBoth GPIO interrupt is asserted at logic
>> level high.
>>
>> Link: https://learn.microsoft.com/en-us/windows-hardware/drivers/bringup/gpio-controller-device-specific-method---dsm-
>> Signed-off-by: Armin Wolf <W_Armin@gmx.de>
>> ---
>>   drivers/gpio/gpiolib-acpi-core.c | 67 ++++++++++++++++++++++++++++++--
>>   1 file changed, 64 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpio/gpiolib-acpi-core.c b/drivers/gpio/gpiolib-acpi-core.c
>> index eb8a40cfb7a9..f7c58ae4d423 100644
>> --- a/drivers/gpio/gpiolib-acpi-core.c
>> +++ b/drivers/gpio/gpiolib-acpi-core.c
>> @@ -15,6 +15,8 @@
>>   #include <linux/irq.h>
>>   #include <linux/mutex.h>
>>   #include <linux/pinctrl/pinctrl.h>
>> +#include <linux/printk.h>
>> +#include <linux/uuid.h>
>>   
>>   #include <linux/gpio/consumer.h>
>>   #include <linux/gpio/driver.h>
>> @@ -23,6 +25,14 @@
>>   #include "gpiolib.h"
>>   #include "gpiolib-acpi.h"
>>   
>> +/*
>> + * Defined by Microsoft at https://learn.microsoft.com/en-us/windows-hardware/drivers/bringup/
>> + * gpio-controller-device-specific-method---dsm-.
>> + */
>> +static const guid_t acpi_gpio_microsoft_guid = GUID_INIT(0x4F248F40, 0xD5E2, 0x499F, 0x83, 0x4C,
>> +							 0x27, 0x75, 0x8E, 0xA1, 0xCD, 0x3F);
>> +#define ACPI_GPIO_DSM_ACTIVE_BOTH_POLARITY 1
>> +
>>   /**
>>    * struct acpi_gpio_event - ACPI GPIO event handler data
>>    *
>> @@ -216,9 +226,11 @@ bool acpi_gpio_get_io_resource(struct acpi_resource *ares,
>>   EXPORT_SYMBOL_GPL(acpi_gpio_get_io_resource);
>>   
>>   static void acpi_gpiochip_request_irq(struct acpi_gpio_chip *acpi_gpio,
>> -				      struct acpi_gpio_event *event)
>> +				      struct acpi_gpio_event *event,
>> +				      union acpi_object *obj)
>>   {
>>   	struct device *parent = acpi_gpio->chip->parent;
>> +	bool active_low = true;
>>   	int ret, value;
>>   
>>   	ret = request_threaded_irq(event->irq, NULL, event->handler,
>> @@ -248,18 +260,67 @@ static void acpi_gpiochip_request_irq(struct acpi_gpio_chip *acpi_gpio,
>>   	if (acpi_gpio_need_run_edge_events_on_boot() &&
>>   	    ((event->irqflags & (IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING)) ==
>>   	     (IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING))) {
>> +		if (obj) {
>> +			for (u32 i = 0; i < obj->package.count; i++) {
>> +				if (obj->package.elements[i].integer.value == event->pin) {
>> +					active_low = false;
>> +					break;
>> +				}
>> +			}
>> +		}
>> +
>>   		value = gpiod_get_raw_value_cansleep(event->desc);
>> -		if (value == 0)
>> +		if ((active_low && value == 0) || (!active_low && value == 1))
>>   			event->handler(event->irq, event);
>>   	}
>>   }
>>   
>> +static union acpi_object *acpi_gpiochip_evaluate_dsm(struct acpi_gpio_chip *acpi_gpio)
>> +{
>> +	struct device *parent = acpi_gpio->chip->parent;
>> +	union acpi_object dummy = {
>> +		.package = {
>> +			.type = ACPI_TYPE_PACKAGE,
>> +			.count = 0,
>> +			.elements = NULL,
>> +		},
>> +	};
>> +	acpi_handle handle = ACPI_HANDLE(parent);
>> +	union acpi_object *obj;
>> +
>> +	if (!acpi_check_dsm(handle, &acpi_gpio_microsoft_guid, 0,
>> +			    BIT(ACPI_GPIO_DSM_ACTIVE_BOTH_POLARITY)))
>> +		return NULL;
>> +
>> +	obj = acpi_evaluate_dsm_typed(handle, &acpi_gpio_microsoft_guid, 0,
>> +				      ACPI_GPIO_DSM_ACTIVE_BOTH_POLARITY,
>> +				      &dummy, ACPI_TYPE_PACKAGE);
>> +	if (obj) {
>> +		for (u32 i = 0; i < obj->package.count; i++) {
>> +			if (obj->package.elements[i].type == ACPI_TYPE_INTEGER)
>> +				continue;
>> +
>> +			dev_err(parent, FW_BUG "Ignoring GPIO _DSM due to invalid data\n");
>> +			ACPI_FREE(obj);
>> +			return NULL;
>> +		}
>> +	}
>> +
>> +	return obj;
>> +}
>> +
>>   static void acpi_gpiochip_request_irqs(struct acpi_gpio_chip *acpi_gpio)
>>   {
>>   	struct acpi_gpio_event *event;
>> +	union acpi_object *obj;
>> +
>> +	obj = acpi_gpiochip_evaluate_dsm(acpi_gpio);
>>   
> So intead of here I think it should be called from
> acpi_gpiochip_request_interrupts() after acpi_walk_resources() and make it
> fill new field active_low in struct acpi_gpio_event.
>
> Then use that field in acpi_gpiochip_request_irq().

Good idea, i will implement this when i find the time.

Thanks,
Armin Wolf

>>   	list_for_each_entry(event, &acpi_gpio->events, node)
>> -		acpi_gpiochip_request_irq(acpi_gpio, event);
>> +		acpi_gpiochip_request_irq(acpi_gpio, event, obj);
>> +
>> +	if (obj)
>> +		ACPI_FREE(obj);
>>   }
>>   
>>   static enum gpiod_flags
>> -- 
>> 2.39.5
>>
>