Re: [PATCH v3 2/2] pinctrl: stm32: Add level interrupt support to gpio irq chip

From: Marek Vasut
Date: Mon Mar 23 2020 - 19:52:43 EST


On 3/23/20 8:49 PM, Marc Zyngier wrote:
> On Mon, 23 Mar 2020 20:37:54 +0100
> Marek Vasut <marex@xxxxxxx> wrote:
>
>> On 3/23/20 8:31 PM, Marc Zyngier wrote:
>>> On Mon, 23 Mar 2020 20:19:39 +0100
>>> Marek Vasut <marex@xxxxxxx> wrote:
>>>
>>>> On 3/23/20 8:04 PM, Marek Vasut wrote:
>>>>> On 2/20/20 10:17 AM, Marc Zyngier wrote:
>>>>>> On 2020-02-20 09:04, Linus Walleij wrote:
>>>>>>> On Wed, Feb 19, 2020 at 3:32 PM Alexandre Torgue
>>>>>>> <alexandre.torgue@xxxxxx> wrote:
>>>>>>>
>>>>>>>> GPIO hardware block is directly linked to EXTI block but EXTI handles
>>>>>>>> external interrupts only on edge. To be able to handle GPIO interrupt on
>>>>>>>> level a "hack" is done in gpio irq chip: parent interrupt (exti irq
>>>>>>>> chip)
>>>>>>>> is retriggered following interrupt type and gpio line value.
>>>>>>>>
>>>>>>>> Signed-off-by: Alexandre Torgue <alexandre.torgue@xxxxxx>
>>>>>>>> Tested-by: Marek Vasut <marex@xxxxxxx>
>>>>>>>
>>>>>>> Reviewed-by: Linus Walleij <linus.walleij@xxxxxxxxxx>
>>>>>>>
>>>>>>> If Marc want to merge it with patch 1/2 go ahead!
>>>>>>
>>>>>> I'll queue the whole thing for 5.7.
>>>>>
>>>>> I have a feeling this doesn't work with threaded interrupts.
>>>>>
>>>>> If the interrupt handler runs in a thread context, the EOI will happen
>>>>> almost right away (while the IRQ handler runs) and so will the code
>>>>> handling the IRQ retriggering. But since the IRQ handler still runs and
>>>>> didn't return yet, the retriggering doesn't cause the IRQ handler to be
>>>>> called again once it finishes, even if the IRQ line is still asserted.
>>>>> And that could result in some of the retriggers now happening I think.
>>>>> Or am I doing something wrong ?
>>>>
>>>> The patch below makes my usecase work, but I don't know whether it's
>>>> correct. Basically once the threaded IRQ handler finishes and unmasks
>>>> the IRQ, check whether the line is asserted and retrigger if so.
>>>>
>>>> diff --git a/drivers/pinctrl/stm32/pinctrl-stm32.c
>>>> b/drivers/pinctrl/stm32/pinctrl-stm32.c
>>>> index 9ac9ecfc2f34..060dbcb7ae72 100644
>>>> --- a/drivers/pinctrl/stm32/pinctrl-stm32.c
>>>> +++ b/drivers/pinctrl/stm32/pinctrl-stm32.c
>>>> @@ -371,12 +371,26 @@ static void
>>>> stm32_gpio_irq_release_resources(struct irq_data *irq_data)
>>>> gpiochip_unlock_as_irq(&bank->gpio_chip, irq_data->hwirq);
>>>> }
>>>>
>>>> +static void stm32_gpio_irq_unmask(struct irq_data *d)
>>>> +{
>>>> + struct stm32_gpio_bank *bank = d->domain->host_data;
>>>> + int level;
>>>> +
>>>> + irq_chip_unmask_parent(d);
>>>> +
>>>> + /* If level interrupt type then retrig */
>>>> + level = stm32_gpio_get(&bank->gpio_chip, d->hwirq);
>>>> + if ((level == 0 && bank->irq_type[d->hwirq] ==
>>>> IRQ_TYPE_LEVEL_LOW) ||
>>>> + (level == 1 && bank->irq_type[d->hwirq] == IRQ_TYPE_LEVEL_HIGH))
>>>> + irq_chip_retrigger_hierarchy(d);
>>>> +}
>>>> +
>>>> static struct irq_chip stm32_gpio_irq_chip = {
>>>> .name = "stm32gpio",
>>>> .irq_eoi = stm32_gpio_irq_eoi,
>>>> .irq_ack = irq_chip_ack_parent,
>>>> .irq_mask = irq_chip_mask_parent,
>>>> - .irq_unmask = irq_chip_unmask_parent,
>>>> + .irq_unmask = stm32_gpio_irq_unmask,
>>>> .irq_set_type = stm32_gpio_set_type,
>>>> .irq_set_wake = irq_chip_set_wake_parent,
>>>> .irq_request_resources = stm32_gpio_irq_request_resources,
>>>>
>>>
>>> OK, I see your problem now.
>>>
>>> The usual flow is along the line of Ack+Eoi, and that's what the
>>> current code guarantees.
>>>
>>> Threaded interrupts do Ack+Mask+Eoi, followed by an Unmask once the
>>> thread finishes. This unmask needs to do the retrigger as well, as you
>>> found out.
>>>
>>> Can you please refactor the above so that we have the common code
>>> between unmask and eoi in a separate function, send a proper patch, and
>>> I'll apply it on top of the current irq/irqchip-5.7 branch.
>>
>> Sure, I can. Do we still need this retriggering in the irq_eoi too ?
>
> Absolutely, because that's what matters for the non-threaded case
> (there is no mask/unmask on that path). It is also never wrong to
> over-resample (it just slows things down).
>
>> Also, are there any other hidden details I might've missed ?
>
> Probably. But let's fix one bug at a time, shall we? ;-) And let's hope
> that ST doesn't take this as a excuse not to clean up their act in
> their next SoC!

Indeed.

Patch is out, thanks for the feedback :)