[PATCH] gpio: lynxpoint: lock IRQs when starting them

[PATCH] gpio: lynxpoint: lock IRQs when starting them

[Linus Walleij]

This uses the new API for tagging GPIO lines as in use by
IRQs. This enforces a few semantic checks on how the underlying
GPIO line is used.

Cc: Mathias Nyman <mathias.nyman@linux.intel.com>
Cc: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
---
 drivers/gpio/gpio-lynxpoint.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/drivers/gpio/gpio-lynxpoint.c b/drivers/gpio/gpio-lynxpoint.c
index f1ca9d5880b6..8c7634250aff 100644
--- a/drivers/gpio/gpio-lynxpoint.c
+++ b/drivers/gpio/gpio-lynxpoint.c
@@ -301,6 +301,24 @@ static void lp_irq_disable(struct irq_data *d)
 	spin_unlock_irqrestore(&lg->lock, flags);
 }
 
+static unsigned int lp_irq_startup(struct irq_data *d)
+{
+	struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
+
+	if (gpio_lock_as_irq(&lg->chip, irqd_to_hwirq(d)))
+		dev_err(lg->chip.dev,
+			"unable to lock HW IRQ %lu for IRQ\n",
+			irqd_to_hwirq(d));
+	return 0;
+}
+
+static void lp_irq_shutdown(struct irq_data *d)
+{
+	struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
+
+	gpio_unlock_as_irq(&lg->chip, irqd_to_hwirq(d));
+}
+
 static struct irq_chip lp_irqchip = {
 	.name = "LP-GPIO",
 	.irq_mask = lp_irq_mask,
@@ -308,6 +326,8 @@ static struct irq_chip lp_irqchip = {
 	.irq_enable = lp_irq_enable,
 	.irq_disable = lp_irq_disable,
 	.irq_set_type = lp_irq_type,
+	.irq_startup = lp_irq_startup,
+	.irq_shutdown = lp_irq_shutdown,
 	.flags = IRQCHIP_SKIP_SET_WAKE,
 };
 
-- 
1.8.3.1

Re: [PATCH] gpio: lynxpoint: lock IRQs when starting them

[Grygorii Strashko]

Hi Linus,

On 11/20/2013 04:42 PM, Linus Walleij wrote:
> This uses the new API for tagging GPIO lines as in use by
> IRQs. This enforces a few semantic checks on how the underlying
> GPIO line is used.
> 
> Cc: Mathias Nyman <mathias.nyman@linux.intel.com>
> Cc: Mika Westerberg <mika.westerberg@linux.intel.com>
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
>   drivers/gpio/gpio-lynxpoint.c | 20 ++++++++++++++++++++
>   1 file changed, 20 insertions(+)
> 
> diff --git a/drivers/gpio/gpio-lynxpoint.c b/drivers/gpio/gpio-lynxpoint.c
> index f1ca9d5880b6..8c7634250aff 100644
> --- a/drivers/gpio/gpio-lynxpoint.c
> +++ b/drivers/gpio/gpio-lynxpoint.c
> @@ -301,6 +301,24 @@ static void lp_irq_disable(struct irq_data *d)
>   	spin_unlock_irqrestore(&lg->lock, flags);
>   }
>   
> +static unsigned int lp_irq_startup(struct irq_data *d)
> +{
> +	struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
> +
> +	if (gpio_lock_as_irq(&lg->chip, irqd_to_hwirq(d)))
> +		dev_err(lg->chip.dev,
> +			"unable to lock HW IRQ %lu for IRQ\n",
> +			irqd_to_hwirq(d));
> +	return 0;
> +}
> +
> +static void lp_irq_shutdown(struct irq_data *d)
> +{
> +	struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
> +
> +	gpio_unlock_as_irq(&lg->chip, irqd_to_hwirq(d));
> +}

Seems, such changes may be risky, because .irq_startup()
and irq_enable()/irq_umask() are mutually exclusive, at least at IRQ request time.
request_threaded_irq()->__setup_irq()->irq_startup()

More over, IRQ core assumes that IRQ is enabled, unmasked and ready for use after
.irq_startup() call. You can check functions irq/chip.c->irq_startup() for more info.

So, .irq_enable() functionality need to be duplicated in .irq_startup() at least.

if you agree - above comment is valid for most of similar recent patches ;)

> +
>   static struct irq_chip lp_irqchip = {
>   	.name = "LP-GPIO",
>   	.irq_mask = lp_irq_mask,
> @@ -308,6 +326,8 @@ static struct irq_chip lp_irqchip = {
>   	.irq_enable = lp_irq_enable,
>   	.irq_disable = lp_irq_disable,
>   	.irq_set_type = lp_irq_type,
> +	.irq_startup = lp_irq_startup,
> +	.irq_shutdown = lp_irq_shutdown,
>   	.flags = IRQCHIP_SKIP_SET_WAKE,
>   };
>   
> 

Regards,
-grygorii

Re: [PATCH] gpio: lynxpoint: lock IRQs when starting them

[Linus Walleij]

On Wed, Nov 20, 2013 at 7:29 PM, Grygorii Strashko
<grygorii.strashko@ti.com> wrote:
> On 11/20/2013 04:42 PM, Linus Walleij wrote:
>>
>> This uses the new API for tagging GPIO lines as in use by
>> IRQs. This enforces a few semantic checks on how the underlying
>> GPIO line is used.
(...)
>> +static unsigned int lp_irq_startup(struct irq_data *d)
>> +{
>> +     struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
>> +
>> +     if (gpio_lock_as_irq(&lg->chip, irqd_to_hwirq(d)))
>> +             dev_err(lg->chip.dev,
>> +                     "unable to lock HW IRQ %lu for IRQ\n",
>> +                     irqd_to_hwirq(d));
>> +     return 0;
>> +}
>> +
>> +static void lp_irq_shutdown(struct irq_data *d)
>> +{
>> +     struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
>> +
>> +     gpio_unlock_as_irq(&lg->chip, irqd_to_hwirq(d));
>> +}
>
> Seems, such changes may be risky, because .irq_startup()
> and irq_enable()/irq_umask() are mutually exclusive, at least at IRQ request time.
> request_threaded_irq()->__setup_irq()->irq_startup()
>
> More over, IRQ core assumes that IRQ is enabled, unmasked and ready for use after
> .irq_startup() call. You can check functions irq/chip.c->irq_startup() for more info.
>
> So, .irq_enable() functionality need to be duplicated in .irq_startup() at least.
>
> if you agree - above comment is valid for most of similar recent patches ;)

Yep. I just showcase what a worthless IRQ core user I am...

Now what would be the best way to call this?

Should I just move this into the .enable callback, and if there is no
such callback, create it and call the .unmask explicitly
at the end of it? It would seem a bit counter-intuitive to do this
in the .mask/.unmask callback, as that should only do exactly
that - mask/unmask.

I'll try this approach...

Yours,
Linus Walleij

Re: [PATCH] gpio: lynxpoint: lock IRQs when starting them

[Grygorii Strashko]

On 11/26/2013 11:52 AM, Linus Walleij wrote:
> On Wed, Nov 20, 2013 at 7:29 PM, Grygorii Strashko
> <grygorii.strashko@ti.com> wrote:
>> On 11/20/2013 04:42 PM, Linus Walleij wrote:
>>>
>>> This uses the new API for tagging GPIO lines as in use by
>>> IRQs. This enforces a few semantic checks on how the underlying
>>> GPIO line is used.
> (...)
>>> +static unsigned int lp_irq_startup(struct irq_data *d)
>>> +{
>>> +     struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
>>> +
>>> +     if (gpio_lock_as_irq(&lg->chip, irqd_to_hwirq(d)))
>>> +             dev_err(lg->chip.dev,
>>> +                     "unable to lock HW IRQ %lu for IRQ\n",
>>> +                     irqd_to_hwirq(d));
>>> +     return 0;
>>> +}
>>> +
>>> +static void lp_irq_shutdown(struct irq_data *d)
>>> +{
>>> +     struct lp_gpio *lg = irq_data_get_irq_chip_data(d);
>>> +
>>> +     gpio_unlock_as_irq(&lg->chip, irqd_to_hwirq(d));
>>> +}
>>
>> Seems, such changes may be risky, because .irq_startup()
>> and irq_enable()/irq_umask() are mutually exclusive, at least at IRQ request time.
>> request_threaded_irq()->__setup_irq()->irq_startup()
>>
>> More over, IRQ core assumes that IRQ is enabled, unmasked and ready for use after
>> .irq_startup() call. You can check functions irq/chip.c->irq_startup() for more info.
>>
>> So, .irq_enable() functionality need to be duplicated in .irq_startup() at least.
>>
>> if you agree - above comment is valid for most of similar recent patches ;)
>
> Yep. I just showcase what a worthless IRQ core user I am...

Yeah. That's the Gray hole in Linux's universe :)

>
> Now what would be the best way to call this?
>
> Should I just move this into the .enable callback, and if there is no
> such callback, create it and call the .unmask explicitly
> at the end of it? It would seem a bit counter-intuitive to do this
> in the .mask/.unmask callback, as that should only do exactly
> that - mask/unmask.
>
> I'll try this approach...

I'm glad to help. Seems you've made it work.


Regards,
-grygorii