Message-ID: <bf5444e4-8f3f-4a56-be67-29857726b119@csgroup.eu>
Date: Fri, 29 Aug 2025 11:41:24 +0200
From: Christophe Leroy <christophe.leroy@...roup.eu>
To: Krzysztof Kozlowski <krzk@...nel.org>, Rob Herring <robh@...nel.org>
Cc: Qiang Zhao <qiang.zhao@....com>, Linus Walleij
 <linus.walleij@...aro.org>, Bartosz Golaszewski <brgl@...ev.pl>,
 Krzysztof Kozlowski <krzk+dt@...nel.org>, Conor Dooley
 <conor+dt@...nel.org>, linux-kernel@...r.kernel.org,
 linuxppc-dev@...ts.ozlabs.org, linux-arm-kernel@...ts.infradead.org,
 linux-gpio@...r.kernel.org, devicetree@...r.kernel.org
Subject: Re: [PATCH v3 5/6] dt-bindings: soc: fsl: qe: Add support of IRQ in
 QE GPIO



Le 29/08/2025 à 11:16, Krzysztof Kozlowski a écrit :
> On 29/08/2025 10:35, Christophe Leroy wrote:
>>
>>
>> Le 29/08/2025 à 09:47, Krzysztof Kozlowski a écrit :
>>> On 28/08/2025 16:12, Christophe Leroy wrote:
>>>>
>>>>
>>>> Le 28/08/2025 à 15:28, Rob Herring a écrit :
>>>>> On Mon, Aug 25, 2025 at 2:20 AM Christophe Leroy
>>>>> <christophe.leroy@...roup.eu> wrote:
>>>>>>
>>>>>> In the QE, a few GPIOs are IRQ capable. Similarly to
>>>>>> commit 726bd223105c ("powerpc/8xx: Adding support of IRQ in MPC8xx
>>>>>> GPIO"), add IRQ support to QE GPIO.
>>>>>>
>>>>>> Add property 'fsl,qe-gpio-irq-mask' similar to
>>>>>> 'fsl,cpm1-gpio-irq-mask' that define which of the GPIOs have IRQs.
>>>>>
>>>>> Why do you need to know this? The ones that have interrupts will be
>>>>> referenced by an 'interrupts' property somewhere.
>>>>
>>>> I don't follow you. The ones that have interrupts need to be reported by
>>>> gc->qe_gpio_to_irq[] so that gpiod_to_irq() return the IRQ number, for
>>>> instance to gpio_sysfs_request_irq() so that it can install an irq
>>>> handler. I can't see where they would be referenced by an "interrupts"
>>>> property.
>>>
>>> They would be referenced by every consumer of these interrupts. IOW,
>>> this property is completely redundant, because DT holds this information
>>> already in other place.
>>
>> But the gpio controller _is_ the consumer of these interrupts, it it
>> _not_ the provider.
>>
>> The interrupts are provided by a separate interrupt controller. Let's
>> take the exemple of powerpc 8xx. Here is the list of interrupts handled
>> by the CPM interrupt controller on the 8xx:
>>
>> 1 - GPIO Port C Line 4 interrupt
>> 2 - GPIO Port C Line 5 interrupt
>> 3 - SMC2 Serial controller interrupt
>> 4 - SMC1 Serial controller interrupt
>> 5 - SPI controller interrupt
>> 6 - GPIO Port C Line 6 interrupt
>> 7 - Timer 4 interrupt
>> 8 - SCCd Serial controller interrupt
>> 9 - GPIO Port C Line 7 interrupt
>> 10 - GPIO Port C Line 8 interrupt
>> 11 - GPIO Port C Line 9 interrupt
>> 12 - Timer 3 interrupt
>> 13 - SCCc Serial controller interrupt
>> 14 - GPIO Port C Line 10 interrupt
>> 15 - GPIO Port C Line 11 interrupt
>> 16 - I2C Controller interrupt
>> 17 - RISC timer table interrupt
>> 18 - Timer 2 interrupt
>> 19 - SCCb Serial controller interrupt
>> 20 - IDMA2 interrupt
>> 21 - IDMA1 interrupt
>> 22 - SDMA channel bus error interrupt
>> 23 - GPIO Port C Line 12 interrupt
>> 24 - GPIO Port C Line 13 interrupt
>> 25 - Timer 1 interrupt
>> 26 - GPIO Port C Line 14 interrupt
>> 27 - SCCd Serial controller interrupt
>> 28 - SCCc Serial controller interrupt
>> 29 - SCCb Serial controller interrupt
>> 30 - SCCa Serial controller interrupt
>> 31 - GPIO Port C Line 15 interrupt
> 
> That list is fixed per soc/device, so already implied by compatible.
> 
>>
>> As you can see in the list, the GPIO line interrupts are nested with
>> other types of interrupts so GPIO controller and Interrupt controller
>> are to be keept independant.
>>
>> That's more or less the same here with my series, patch 1 implements an
>> interrupt controller (documented in patch 6) and then the GPIO
>> controllers consume the interrupts, for instance in gpiolib functions
>> gpio_sysfs_request_irq() [drivers/gpio/gpiolib-sysfs.c] or
>> edge_detector_setup() or debounce_setup() [drivers/gpio/gpiolib-cdev.c]
>>
>> External drivers also use interrupts indirectly. For example driver
>> sound/soc/soc-jack.c, it doesn't have any direct reference to an
>> interrupt. The driver is given an array of GPIOs and then installs an
>> IRQ in function snd_soc_jack_add_gpios() by doing
>>
>> 	request_any_context_irq(gpiod_to_irq(gpios[i].desc),
>> 					      gpio_handler,
>> 					      IRQF_SHARED |
>> 					      IRQF_TRIGGER_RISING |
>> 					      IRQF_TRIGGER_FALLING,
>> 					      gpios[i].name,
>> 					      &gpios[i]);
> 
> 
> External drivers do not matter then. Your GPIO controller receives
> specific interrupts (that's the interrupt property) and knows exactly
> how each GPIO maps to it.
> 

Do you mean then that the GPIO driver should already know which line has 
an interrupt and which one doesn't ?

The interrupts are fixed per soc, but today the GPIO driver is generic 
and used on different SOCs that don't have interrupts on the same lines. 
And even on the given SOCs, not all ports have interrupts on the same 
lines. Should all possibilities be hard-coded inside the driver for each 
possible compatible ? The property 'fsl,qe-gpio-irq-mask' is there to 
avoid that and keep the GPIO driver as generic as possible with a single 
compatible 'fsl,mpc8323-qe-pario-bank' that is found in the DTS of 8323 
but also in DTS of 8360, in DTS of 8569, etc... :

arch/powerpc/boot/dts/fsl/mpc8569mds.dts: 
             "fsl,mpc8323-qe-pario-bank";
arch/powerpc/boot/dts/fsl/mpc8569mds.dts: 
             "fsl,mpc8323-qe-pario-bank";
arch/powerpc/boot/dts/kmeter1.dts: 
     "fsl,mpc8323-qe-pario-bank";
arch/powerpc/boot/dts/mpc832x_rdb.dts: 
compatible = "fsl,mpc8323-qe-pario-bank";
arch/powerpc/boot/dts/mpc836x_rdk.dts: 
"fsl,mpc8323-qe-pario-bank";
arch/powerpc/boot/dts/mpc836x_rdk.dts: 
"fsl,mpc8323-qe-pario-bank";

Do you mean we should define one compatible for each of the ports of 
each soc, and encode the mask of interrupts that define which line of 
the port has interrupts in the data field ?

Something like:

static const struct of_device_id qe_gpio_match[] = {
	{
		.compatible = "fsl,mpc8323-qe-pario-bank-a",
		.data = (void *)0x00a00028,
	},
	{
		.compatible = "fsl,mpc8323-qe-pario-bank-b",
		.data = (void *)0x01400050,
	},
	{
		.compatible = "fsl,mpc8323-qe-pario-bank-c",
		.data = (void *)0x00000084,
	},
	{
		.compatible = "fsl,mpc8323-qe-pario-bank-d",
		.data = (void *)0,
	},
	{
		.compatible = "fsl,mpc8360-qe-pario-bank-a",
		.data = (void *)0xXXXXXXXX,
	},
	{
		.compatible = "fsl,mpc8360-qe-pario-bank-b",
		.data = (void *)0xXXXXXXXX,
	},
....
	{},
};
MODULE_DEVICE_TABLE(of, qe_gpio_match);

That would be feasible but would mean updating the driver each time a 
new SOC is added.

Do you mean it should be done that way ?

Isn't the purpose of the device tree to keep drivers as generic as 
possible ?




>>
>>>
>>>>
>>>>>
>>>>>> Here is an exemple for port B of mpc8323 which has IRQs for
>>>>>
>>>>> typo
>>>>>
>>>>>> GPIOs PB7, PB9, PB25 and PB27.
>>>>>>
>>>>>>            qe_pio_b: gpio-controller@...8 {
>>>>>>                    compatible = "fsl,mpc8323-qe-pario-bank";
>>>>>>                    reg = <0x1418 0x18>;
>>>>>>                    interrupts = <4 5 6 7>;
>>>>>>                    interrupt-parent = <&qepic>;
>>>>>>                    gpio-controller;
>>>>>>                    #gpio-cells = <2>;
>>>>>>                    fsl,qe-gpio-irq-mask = <0x01400050>;
>>>>>>            };
>>>>>
>>>>> You are missing #interrupt-cells and interrupt-controller properties.
>>>>
>>>> The gpio controller is not an interrupt controller. The GPIO controller
>>>> is brought by patch 1/6 and documented in patch 6/6.
>>>
>>> Then the IRQ mask property is not right here. If you say "this GPIOs
>>> have IRQs" it means this is an interrupt controller.
>>
>> The mask tells to the GPIO controller which GPIO line has an interrupt
>> (so it can install the edge detector) and which doesn't have an
>> interrupt. The "interrupts" property gives a flat list of interrupts,
>> the mask in the above example tells: interrupt 4 is for line 7,
>> interrupt 5 is for line 9, interrupt 6 is for line 25, interrupt 7 is
>> for line 27. Other lines don't have interrupts.
>>
>>>
>>> If you say this is not an interrupt controller, then you cannot have
>>> here interrupts per some GPIOs, obviously.
>>
>> It has been working that way on powerpc 8xx for 8 years, since commit
>> 726bd223105c ("powerpc/8xx: Adding support of IRQ in MPC8xx GPIO")
>>
>> I don't understand why you say you cannot have
>> here interrupts per some GPIOs. What am I missing ?
> 
> I used conditional. English conditional. If you claim this GPIO
> controller is not an interrupt controller, then obviously it is not an
> interrupt controller and cannot be used as interrupt controller.
> 
> If you use "foo" as interrupt controller, then clearly foo is an
> interrupt controller.
> 

Ah ! ok. I didn't read it like that, sorry.

Thanks
Christophe

Powered by blists - more mailing lists